/hdg/ - Stable Diffusion

Anime girls generated with AI

Index Catalog Archive Bottom Refresh
Name
Options
Subject
Message

Max message length: 12000

files

Max file size: 15.00 MB

Total max file size: 50.00 MB

Max files: 4

Supported file types: GIF, JPG, PNG, WebM, OGG, and more

E-mail
Password

(used to delete files and posts)

Misc

Remember to follow the Rules

The backup domains are located at 8chan.se and 8chan.cc. TOR access can be found here, or you can access the TOR portal from the clearnet at Redchannit 3.0.



8chan.moe is a hobby project with no affiliation whatsoever to the administration of any other "8chan" site, past or present.

>>28238 But anon, I'm on Comfyui :3 Jokes aside, I edit my gens slightly on a outside editor, so metadata gets striped either way. I'd share the OG catboxes if asked though
(3.88 MB 2128x1456 catbox_1ud875.png)

(988.13 KB 1000x1000 00014-2737346106.png)

(968.31 KB 1000x1000 00011-2937582114.png)

(3.92 MB 2128x1456 00089-20836112.png)

>>28042 >asuka
(2.50 MB 1248x1824 catbox_y07r3i.png)

(2.18 MB 1248x1824 catbox_2lc6dg.png)

(2.47 MB 1248x1824 catbox_jz5zlc.png)

(1.68 MB 1248x1824 catbox_7jp3m7.png)

New bread, more beach wives
Huh, does Epsilion 1.1 know who Niyaniya is? She's an extremely recent character and I got this by just prompting for professor_niyaniya_(blue_archive) , blonde
>>28242 That's actually extremely useful, thanks anon These images weren't saved using it, but I'll use it going forward
>>28243 Also, some of these illu finetunes seem to know who Yunli is without Loras pretty well. Yunli's very first picture was uploaded to danbooru in June so I think some of them are using a dump from around July/August.
>>28243 >>28245 Noob used up-to-date Danbooru as of the time it started training. So about mid-late October.
>>28246 I don't think it's october, it doesn't recognize characters like Lingsha and it barely understands Niyaniya so it feels more like an August dump.
>>28247 Remember that it takes time for characters to get enough fanart to be recognized.
>>28248 Yeah, it's why I assume it's somewhere in July or August. Niyaniya became a hotness in BA and got art pretty quickly so by October, the model should know her as well as Yunli given that she has more pics than her. August seems like the safest bet.
Nahida sex
>>28237 catboxes?
(2.93 MB 1248x1824 catbox_hhn749.png)

(2.72 MB 1248x1824 catbox_ybbmw4.png)

(2.65 MB 1248x1824 catbox_epdvuq.png)

(2.21 MB 1248x1824 catbox_tu9evd.png)

Testing DPM++ 2M again(I have no idea what samplers are the best anymore).
is vpred suppose to be bad at backgrounds? vpred is insane, but the way u have to play with different cfg scales for different artists is exhausting when u have shitty gpu
any opinions on 0.9R?
>>28257 not bothering with it because their last two shitmixes were ass and I don't care about whatever weird realism shit they're doing. hoping base 0.9 and 1.0 aren't fucked over by realism data.
>>28258 0.75 is feeling pretty complete. this is a character with 106 tags and a complex costume and it got her like 70% right.
really my biggest disappointment left with the model is it still has a lot of resolution-dependent torso stretching
>>28259 >0.75 is feeling pretty complete styles could use some improvement
>>28258 Why can't they just stick to making a pure anime model first and do experiments later, jeez.
>>28257 .9R is handing out free watermelon slices.
>>28264 >pure anime model first did you miss the part where they added e621 right from the start? lol.
>>28266 I thought that was after eps 0.5.
>>28267 no, after 0.5 they trained text encoder which caused some issues in 0.75 e621 was part of the noob dataset right from the start. illustrious is the only pure danbooru
(2.42 MB 1248x1824 00042-325698705.png)

(2.20 MB 1248x1824 00053-325698707.png)

(2.45 MB 1248x1824 00055-325698709.png)

(2.19 MB 1248x1824 00058-325698712.png)

(2.76 MB 1248x1824 catbox_3d9fch.png)

(2.77 MB 1248x1824 catbox_cqc018.png)

(2.27 MB 1248x1824 catbox_087ogq.png)

(2.63 MB 1248x1824 catbox_saecvs.png)

>>28269 cute
(2.31 MB 1248x1824 00070-1014647082.png)

(2.52 MB 1248x1824 00065-1415891815.png)

(2.41 MB 1248x1824 00059-1407792194.png)

(2.74 MB 1248x1824 00063-1933927739.png)

>>28270 thanks! haven't proompted since the start of the year, was a trip reading up on all the new shit. my extensive sd1.5 lora collection is now useless, but the noob ai model comprehension is impressive.
anyone have 0.75 vs 0.9R comparison grids? i was just made aware the latter is not a slopmix and apparently fine despite the introduction of more realistic data, and I can't run tests myself for a bit.
>>28271 catbox for 1 and 4 pls?
>>28271 Welcome backy!
(2.49 MB 1248x1824 catbox_wwhlcc.png)

(2.40 MB 1248x1824 catbox_jhnwhi.png)

(2.47 MB 1248x1824 catbox_g01hjy.png)

(2.36 MB 1248x1824 catbox_tg7y7q.png)

Stupid sexy Christmas brat.
>>28271 catboxes?
>>28273 >>28277 guess I need to install the 8chan hdg extension again https://files.catbox.moe/on8x65.png https://files.catbox.moe/8oajun.png https://files.catbox.moe/hwzj9x.png https://files.catbox.moe/az2wtr.png any prompt tips for the new models are welcome >>28276 hnnng
(284.99 KB 1248x1824 catbox_fy9n50.jpg)

(209.37 KB 1248x1824 catbox_pnk4sj.jpg)

(297.04 KB 1824x1248 catbox_3tbxj9.jpg)

(2.84 MB 1824x1248 catbox_dpa71j.png)

(2.49 MB 1248x1824 catbox_gfudtr.png)

(2.46 MB 1248x1824 catbox_chw8g4.png)

(2.39 MB 1248x1824 catbox_c5lmmh.png)

(2.42 MB 1248x1824 catbox_jw5y4l.png)

time to cook a new prompt
(1.09 MB 1664x2432 catbox_imq4b1.jpg)

(1.41 MB 832x1216 catbox_0qwwjq.png)

(1.57 MB 832x1216 catbox_auv9as.png)

NAI's artist:rustle produces a better result than NoobAI's rustle. Every time I use rustle with NoobAI the image comes out dark and a bit sepia.
>>28281 Can't check meta right now, are you using 0.6 or 0.75? I had similar issues with muk on 0.6 but 0.75 fixed them.
I keep going from prompting the most degenerate and fucked up /d/ shit I can think of to just generating basic anime girls and inpainting them naked
>>28281 imo noob's artist tags just aren't as good as nai's but loras seem to fully compensate for that. honestly noob with style loras feels much better to use in general, shame that there's not much good ones around right now (understandable why though)
>>28281 Have you tried "sepia" in the negative prompt?
>>28248 decades in some cases (╥﹏╥)
>>28284 I really wish I could find more style loras but it seems people hardly share them, at least on hdg and vt
>>28287 lot of trainers are waiting for things to stabilize
>>28290 very nice
>>28290 catboxes?
>>28202 >>28221 >>28223 I'm quite late but DAMN any catboxes? Please.
(299.88 KB 832x1216 catbox_8xsam2.jpg)

(327.18 KB 832x1216 catbox_pgyuof.jpg)

>>28282 using 0.9R. let me know if rustle looks ok on 0.75. >>28285 sepia in the negs does help a little. but it's still there and way more pronounced than on NAI. picrel rustle vs no rustle >>28284 I'm the anon that trained the rustle lora on here long ago. if I figure out how to train loras on SDXL I can take another crack at it. haven't touched loras at all with the new models yet. >>28283 a familiar cycle. purge degeneracy with cute stuff, gradually get lewder, repeat.
(2.41 MB 1248x1824 catbox_462tty.png)

>>28296 yeah this is peak SEXO >>28297 very cute
Has anyone made a lora of this cute boat? I saw one for Pony but nothing on Illu. I quit AL ages ago but that doesn't stop me from loving cute white haired little girls. https://civitai.com/models/430866/flandre-xl-azur-lane
>>28300 cute cat girls are how I first got into loli
has anyone been able to get kuroboshi kouhaku to work well on vpred? strange that I cant seem to get an "accurate" depiction when he has 800 entries
>>28303 the image quality here is a little..... grainy jokes aside it's fitting, very cute
(451.77 KB 1248x1824 catbox_ynru9j.jpg)

(430.33 KB 1248x1824 catbox_i5ac2p.jpg)

(448.72 KB 1248x1824 catbox_jmhan5.jpg)

(438.77 KB 1248x1824 catbox_vgfp4g.jpg)

>>28306 cunnyfication is based
>>28284 If you use a less trained version of Noob instead of 1.0, the Danbooru information shines through better. So basically, it's better at Japanese artists, worse at western e621 artists. I was having good results with Epsilon 0.75 but even lower might be better, haven't tried those. The thing to keep in mind is that more training doesn't necessarily mean better, the Danbooru information is getting diluted by the e621 information. Whether that is a good thing or not depends on how much you like e621 shit, most of it is garbage in my opinion.
(402.51 KB 1248x1824 catbox_7to0ui.jpg)

(356.34 KB 1248x1824 catbox_vwl5bh.jpg)

(396.13 KB 1248x1824 catbox_c7b7bd.jpg)

(378.75 KB 1248x1824 catbox_xc1h91.jpg)

>>28307 truly
>>28308 I actually was mostly talking about vpred versions. I only tried 1.0 when it comes to eps and even there artist tags felt kinda stronger compared to vpred... I'm satisfied with vpred+lora for now though, I like the benefits it provides despite all the jank. Maybe I should try pure illustrious or eps0.5 some time though.
why are furfags so fucking worthless? why do they need to give out their opinion or even simply reply to EVERY single question ESPECIALLY when they know NOTHING about the subject? both the illustrious and anlatan discords are FULL of them and they all act the fucking same, you can't ask ANYTHING without them butting in with their worthless fucking non-replies AND THEY GO ON AND ON AND ON, they literally can't stop blabbering once they start replying holy FUCK total furfag death can't come soon enough
>>28308 >the Danbooru information is getting diluted by the e621 information is this how it works? shouldnt sdxl be able to handle 12 million images without getting "diluted"
>>28312 It's a 6GB "brain." That's not enough to store 12 million images' worth of data without losing massive amounts of information. Think about it this way, if you have a dataset of exactly 1 picture, the only thing the model will be able to produce is that one picture, but it will reproduce it perfectly. Now extrapolate that to an entire website. A dataset of only Danbooru will be able to reproduce the "soul" of Danbooru perfectly. A dataset of only e621 will reproduce the "soul" of e621 perfectly. A dataset of both will be half as good at both. I don't know what size of "brain" you would need to handle both with no issue, but probably way, way bigger than 6GB. A dataset dump of Danbooru alone is many TB of pictures, not GB. As more evidence of the limits of SDXL, there's not a single checkpoint that can handle both 2D and realism, so western VS eastern art is probably a similar dilemma, although the gap is maybe not as wide.
>>28313 all in how you handle it
v4 preview is out, a chaste model like last time. character segmentation feels like it's about 20% on the mark vs like 3% for all previous models. artist tags are still in. unsurprisingly feels underbaked, probably not worth paying for presently.
>>28316 this is how the segmentation is done, a universal prompt at the top, boxes for character stuff, and a five by five x/y grid for roughly where they should be
Kinda hard to get shit going for v4 considering they probably pruned everything nsfw for the preview
>>28317 So a static grid regional prompter basically?
>>28319 if you choose to, it's optional
>>28316 >v4 preview is out, a chaste model like last time did they have v3 preview? was it considerably worse than release v3? the update blog only mentions v2 and then v3 release after that
>>28321 v1 had a curated version way back in the day that was also bad. pretty much same scenario
>>28322 compared to the full version curated v1 was pretty good (at least in terms of baseline quality) i still unironically think it would've been better to use that as a base model and train the NSFW concepts back in
Haven't prompted in a while, what's the latest thing we're using now? Is it still A111? Forge? Comfy?
>>28327 i'm using reforge, haven't cared to check what forge is up to for a while
>>28328 Where do I get that?
>>28330 Thanks
try invoke, it's really good no they don't pay me no i don't use local no i don't mean the paid service
>>28332 Any reasons why you think it's good?
>>28333 it's not gradio, you don't need to switch tabs constantly, a lot of auto extensions are built-in, you've got nodes to waste time with in case you're one of those dunning-kruger types, the inpainting is better than auto and the tiled canvas+regional guidance are much better bandaids than auto/comfy regional extensions i haven't used it in a long while because i don't use local anymore, i shill it just because i think people are really sleeping on it and it could lead to better/faster/less frustrating workflows yet nobody will even give it a try because "muh ironic anti-corpo time is money i can't waste 2 mins clicking through an installer and 3 mins learning how shit works" shitposters think they're switching from 3dsmax to blender
>>28334 telemetry status?
>>28335 disabled via launch arguments
>>28332 why do people even get triggered by invoke suggestions i tried it some time ago and i think its indeed good, im just more used to a1111 shit and went back to reforge, but invoke honestly looks better thought out to me
>>28337 it seems alright but i bounced off it earlier for lack of obvious equivalents for shit i like i'll try it again tomorrow
>>28337 pick one or more a) shitposters b) (see above) ui consolewar faggottry (moot point because invoke switched to nodes quite some time ago) c) "but i dont want to learn a new ui" (also moot for obvious reasons) d) they think people mean the paid service or that you have to pay (heresy) for the ui or something e) it lacks some niche extension they use (fair although the major/important ones are built-in) d) they think it's slower (probably is?, i doubt it's as fast/optimized as reforge) one shitposter in particular really dislikes the emphasis syntax (appending +/– instead of parentheses) but i find it way more convenient, even using the auto shortcut you still have do highlight manually/by double-clicking and then you have to use two more keys, this way you only have to click and smash +/- my only gripe with it is that it still doesn't have NAI and autocomplete extensions, if it did i would delete my old auto install (which i only use for the NAI extension) in a heartbeat
>>28339 (i'm subscribed to their yt channel and every time i watch their update videos i seethe at how bad the gradio uis and NAI site are)
(4.44 MB 3464x1872 Untitled-1.png)

(4.41 MB 3480x1884 Untitled-2.png)

I'm back, what did I miss >check /hdg/, picrel Looks like I didn't miss a thing
(2.55 MB 1248x1824 catbox_ouzjps.png)

(2.55 MB 1248x1824 catbox_xjswu9.png)

(2.47 MB 1248x1824 catbox_7lbma8.png)

(2.54 MB 1248x1824 catbox_7w15l6.png)

>>28342 >Looks like I didn't miss a thing Yes, you are correct.
>>28308 >>28313 care to quantify any of these wild speculations? This sounds like another "fur suit eyes" situation
>>28344 >source??? read up on sdxl maybe?
call me when v4 is done baking i'm gonna play bloodborne remastered
>>28346 >bloodborne remastered you emulating the game or something?
>>28347 i hate being pedantic but i'm almost positive it's a translation layer instead of actual emulation but yeah
>>28348 I see. I've been in the mood to replay Bloodborne but the game deserves better performance than the shit one you get on a PS4.
>>28349 you need a few mods and bolt-on fixes but people got it working at 60 fps the moment it got in-game with acceptable visuals months ago progress has been lightning-fast, there are very few unresolved issues left
>>28350 Aight, I guess I can wait some more. Gotta beat SMTV and enjoy Elin some more anyway.
>>28351 it's already playable from start to finish including the dlc if you don't want to wait, the only real issues are vertex explosions (fixable by disabling facegen for custom faces, not really a huge loss unless you like making oblivion-level abominations for the memes) and i think one last memory leak on low ram/vram configs (irrelevant for SD users) although i guess you might as well wait until the remaster project hits 1.0, 0.8 was released like two days ago
>>28352 >it's already playable from start to finish including the dlc That's really cool, I'm tempted to play it right now but I'll wait for 1.0.
>>28353 probably won't take too long, guy is zooming through it despite the fact that he's also fixing normal maps and fixing + making bump maps for every texture that needs them (BB has quite a few bump maps already but 95% of them are disabled because they completely fucked up the contrast and ended up looking like shit - meanwhile the b-team made AND nailed every single one for the best souls 2 dlc but left them disabled due to performance issues) bWFnbmV0Oj94dD11cm46YnRpaDpBNjVGQjk0QzBEQzhEN0Q5ODcxNDdCNkY1MzhFNTk2MUYyODkxQUM2JnRyPWh0dHAlM0ElMkYlMkZidDQudC1ydS5vcmclMkZhbm4lM0ZtYWduZXQmZG49JTVCUFM0JTIwRXhjbHVzaXZlJTVEJTIwQmxvb2Rib3JuZSUyMEdhbWUlMjBvZiUyMHRoZSUyMFllYXIlMjBFZGl0aW9uJTIwKENvbXBsZXRlJTIwRWRpdGlvbiklMjAlNUJFVVIlMkZSVVMlNUQlMjAodjEuMDkp
>>28354 i said bump maps but it's displacement maps whatever, you know what i mean
(2.20 MB 1824x1248 catbox_at0lbl.png)

>>28354 Oh shit, thanks.
>>28356 ah fuck now i'm upset, traditional media tags are the first style tags i tried on the v4 preview and they don't do anything i tried messing around with the quality tags and uc presets and they kinda start working (need to mess with it more) but the quality drops significantly as soon as i switch to the light preset and when they do work they're NOWHERE near as good as v3's really hope it's just a dataset issue with the preview or just the preview itself being underbaked i NEED my {traditional media}, marker (medium), {{millipen (medium)}} kino
(4.43 MB 6384x6483 catbox_fwxdqu.jpg)

(3.87 MB 5776x6460 catbox_mvk185.jpg)

(4.07 MB 5776x7123 catbox_z5cga8.jpg)

What's the verdict on vpred v1.0?
>>28360 >>28361 what's the purpose of cyberfix? what does it fix?
>>28362 Cyberfix is just NoobAI + cyberrealisticXL - base sdxl, then noob CLIP added back. Itercomp is the same thing but with itercomp model instead of cyberrealistic. Not sure about the specifics of how perpendicular is different but I believe it's just using the same workflow as the itercomp merge posted last thread >>27890 The slerp merge of the itercomp and cyberfix models I am unsure about, but I did a 50/50 sum in reforge/a1111 and the result was about the same, which is the last comparison in >>28361
>>28364 holy fucking kino that's the best thing i've seen in a long time
Now that 1.0 is out, does anyone have loras planned?
>>28368 I'm kinda burnt out after weeks of iterative updates but I'll likely continue to train on 0.75 if I do more since it works best for me. Unless one of the experimental models is really good, whenever they clean up their HF and public those.
iterslop test prompts, 30 step dpmpp 2m sgm uniform 5cfg/0.7 rescale no adetailer https://files.catbox.moe/1tf646.png https://files.catbox.moe/7rlttk.png https://files.catbox.moe/7rlttk.png https://files.catbox.moe/ydliz6.png imo, 0.75 > 1.0 by a decent amount. 1.0 feels crustier without any real improvement in coherence and the colors are more likely to explode the moment you turn off rescale.
Merry Christmas you degens
Did anyone ever make an updated lora training guide? I seem to be shit at training on SDXL
>>28364 Nice!
(2.86 MB 1248x1824 file.png)

(2.80 MB 1248x1824 file.png)

Merry Christmas
>>28379 cute
>>28393 Impossible to answer this question without training params
>>28394 https://files.catbox.moe/ztuujc.toml AdamW8bit cosine, unet 4e-4, tenc 1e-4, 5000 total steps
>>28395 4e-4 feels too high https://files.catbox.moe/58hkrw.toml this is what I've been using, the specific bake was 70 images so do the math to figure out how many epochs/repeats you need to reach the same step count
>>28395 I would try training without the scaling v pred loss arg. Training the tenc is also useless and potentially harmful, don't do it. My settings are pretty close to >>28396, except on the contrary I'm training with a fairly high learning rate like yours (5e-4), and I also cranked the dim to 64 because I was getting underbaked shit otherwise too.
>>28388 catbox?
my wife needs more art :(
>Enable “Scale v-pred loss" >epoch 20 is underbake >Disable "Scale v-pred loss" >epoch 12 is well done WTF is this option
>>28406 momo wasn't toki your wife?
>default settings >no prompt whatsoever >v4 sfw preview generates the equivalent of "foot focus, soles" surprisingly often incredibly, no, UNFATHOMABLY keyed
plappies go in here
>>28410 wew #3
>>28414 >pose borrowed just tags or i2i?
>>28408 not me but my friendfu is grey wolf. I used toki a lot for test gens cuz she has a very striking design.
>>28416 ah, i see where the fuck is momoura then
Does anyone have a guide or config for training a lora for sdxl models? I'm getting back into it for the first time since the 1.5 era and figured vpred would be the best time for it
>>28412 Yeah I really like that gen as well and I'm probably gonna spend a little bit of time improving it.
i don't have anything witty to say this time i'm just waiting for v4 full
>>28421 >>28422 Now this is some good shit Thanks
>>28428 uooooh
i accidentally the sfw model
Anyone have more of Andrea Davenport from the ghost and molly mcgee? I really like her design and colors but it's hard to find good gens of her
>>28417 i am he...
>>28433 oh did you mean "not mine" instead of "not me"?
cnuuy still can't believe it was that easy to trick it into making bare cunny
(2.76 MB 1344x1728 catbox_qa91la.png)

>>28434 I meant I wasn't the manatee poster but I realized I worded that poorly retard to ESL communication
>>28437 looking back it still kinda makes sense but the "but" threw me off anyway you know what i'm gonna say, hand them over
>>28438 https://files.catbox.moe/xemimf.png idk if I ever uploaded this one but it's very obviously WIP, I burnt out for a while, came back, and am still kinda burnt out just playing with my personal cards from time to time
>>28439 A CHRISTMAS MIRACLE
(1.02 MB 896x1152 00240-1948022297.png)

(1014.32 KB 896x1152 00239-4024696847.png)

(1.07 MB 896x1152 00248-2773669032.png)

(950.67 KB 896x1152 00253-4216235749.png)

(2.82 MB 1344x1728 00021-3376216359.png)

(2.67 MB 1344x1728 00023-1843350828.png)

I fucking love this model
>>28442 ARE WE BRINGING JLULLABY BACK?
Is there any way to do regional/tiled prompting on forge? I remember there was a way to prompt something on a specific region of the image then something different on another region but I forgot how to do it or what was the extension for it
(1.21 MB 896x1152 00204-1055337904.png)

>>28444 Funny how you say that because it's actually angelauxes, not jlullaby
>>28446 >angelauxes I see it now but the 3D-ish shading/effect reminded me more of JL
>>28426 catbox for the first one? or artist mix, love that kind of cunny face
(2.21 MB 2816x8392 catbox_ngqo16.jpg)

(2.60 MB 2816x8392 catbox_x1lpu6.jpg)

I thought I don't have to do this anymore, turns out I was wrong
>>28453 >>28454 great, now i'm thinking about malo and his mart
NOBODY CARES BUT I'M SEVERELY SLEEP-DEPRIVED, I DRANK 2 CANS OF MONSTER AND I'M CURRENTLY BUILDING THAT FUCKING RIG FUCK CORSAIR AND FUCK CASES I SHOULD'VE BOUGHT A BENCH FRAME i'm only doing it because i gotta claim the quixel megascans before epic decides to charge for them
>you have to sign in >you have to send email shit to check you're not a bot >captcha captcha captcha >you gotta dance naked in front of camera to prove you're not a schizo >ip range blocked internet has seriously went downhill lately
>>28457 you forgot the part where the captcha is harder to read than fucking amongus indonesian runes
>>28457 >you gotta dance naked in front of camera to prove you're not a schizo kek
modular psu connectors are the fucking devil
(1.22 MB 1920x1088 catbox_bdd2se.png)

>>28451 nuked the folder. this krisbox is all that remains
(2.07 MB 1920x1088 00052-1289952162.png)

(1.69 MB 1920x1088 00051-1289952161.png)

(1.71 MB 1920x1088 00048-1891041041.png)

(1.99 MB 1920x1088 00045-555704230.png)

(2.05 MB 1920x1088 00053-1289952163.png)

(1.71 MB 1920x1088 00050-1891041043.png)

(1.72 MB 1920x1088 00046-1891041039.png)

(1.58 MB 1920x1088 00056-4239113545.png)

>>28460 works good.
happy new year
(1.63 MB 4080x2517 b.jpg)

(2.95 MB 4080x2288 a.jpg)

>drawfags now: "take up a pencil!!!!!" >drawfags 2021:
>>28469 Both the "pick up a pencil" and "artists are finished" crowds are fucking retarded and most of the people saying both aren't even artists to begin with. Good artists make good work using their available tools to the fullest extent. FDY's style is unique specifically because he draws over 3D renders—which he makes—to get that soft lighting, in no way does that devalue the end result. And it's the same for people who got into art *because* of AI, I'd consider people like brittle to be legitimate artists at this point because they do so much redrawing and inpainting that it's merely AI-assisted.
>>28470 don't think anyone who can't draw/render without ai could be called an artist not that "artist" itself is the highest praise imaginable
(3.36 MB 3696x11015 catbox_35va7v.jpg)

(3.35 MB 3344x11103 catbox_321go2.jpg)

>>28470 I agree, the more you care about gen quality the more time you spend in photoshop I'm annoyed by the "pick up a pencil" crowd as they are not triggered by low quality art or "wasted potential" but are categorically against generative AI from a moral standpoint, in fact they became visibly more desperate after the release of NAIv3
>>28472 Can you link me the loras please
>>28469 funny shitpost aside is that how tianglingdingdongchingchong gets stuff done? that's pretty cool
>>28470 i really can't describe how much contempt i have for the sort of dipshit who tries to resolve their imposter syndrome by playing social games harder rather than practicing harder all forms of art are just rotten with them these days
personally i find both the "artists are finished" shitposters and actual believers to be based (and that includes myself, obviously)
(2.81 MB 1456x1872 catbox_536cwn.png)

(2.84 MB 1456x1872 catbox_yewtev.png)

(2.99 MB 1456x1872 catbox_3j6wyw.png)

(2.73 MB 1456x1872 catbox_n6s1zn.png)

Year of the snek coming.
(1.53 MB 1200x1600 00422-1239516851.png)

(1.58 MB 1200x1600 00415-1904694093.png)

I love cosplay \(cosplay\)
>>28476 Did you edit the second pic in unconventional ways? It has different pixel values from the original file when loaded into a Firefox canvas which fucked up metadata detection For anyone using Firefox, disable color management or the userscript here or on 4chan may not work properly in about:config set gfx.color_management.mode = 0
(616.65 KB 1536x2560 recreate.jpg)

>>28481 It's actually eerie how close it gets to just recreating the training data if you copy and paste the training prompt for one of the pictures. Maybe I shouldn't be surprised.
>>28483 >if you copy and paste the training prompt for one of the pictures this can happen somewhat regularly on base models too if you copy the prompt 1:1 but it's obviously more common if not guaranteed on loras if i'm bored to death i just grab the tags from a db post, plug them in 1:1 and see if it spits out the same image, it's fun for some reason
>>28453 >>28454 Meeth Crym (Final Fantasy Crystal Chronicles: Ring of Fates) NoobAI vpred 1.0 Trigger "meeth, lilty" https://files.catbox.moe/2x8afe.safetensors
>>28457 the spam problem must be getting bad at 4chinks because my old PC i've been using to shitpost can post without problems but my new PC literally sitting right next to me needs to wait 15 minutes then verify email because i'm blocked from image posting if I dont
>>28457 >>28487 i'm blaming /degen/ lol. they don't shy away from posting "I want to lick that old man ass" or realistic gens and videos. sure they get banned and posts deleted,but they don't care.
>>28488 I thought it's related to /pol/ mostly but anyway I blame the jannies. This 15 minute shit does nothing to prevent actual insane schizos from shitting up the boards and only lowers participation from casual anons. Nobody wants to use their email on fucking 4chan and those 15 minute timers are incredibly annoying considering they get reset if you don't post in *exactly* 15 minutes
>>28486 i like her. it's a great lora too. >>28489 it's always the jannies fault tbh.
w-what happened?
>>28492 My understanding is that it started to get a little too good at realism so they had to regex some tags to 3D ones, there was backlash over that and now kurumuz is talking about retraining the model from scratch.
>>28493 i searched kurumuz's messages and apparently "some realistic data got sneaked into the dataset", whatever that means i really hope that doesn't mean losing/nerfing the realism-related tags altogether worst i've seen is some really bad and uncanny serena from /v/ shame though, the update definitely had better prompt adherence and consistency (pic unrelated though this was also quite a bit more consistent on the update)
Anyone know a decent guide on how to make outpainting not suck ass?
>>28492 "asian -photo_(medium) -real_life" still has untagged photo content.
>>28498 thanks for clarifying
>>28499 Just speculation.
(3.62 MB 1664x2432 2379_125839778_p2.png)

(2.19 MB 5376x4032 catbox_riigss.jpg)

Those huge repetitive AI dumps on sad panda serve one exact purpose: what they include are evidently what can be done reliably in AI
(1.70 MB 1248x1824 00014-1609425371.png)

(906.69 KB 1248x1824 00026-133195248.png)

(1.34 MB 1248x1824 00031-3224072633.png)

(1.55 MB 1248x1824 00041-2923310545.png)

>>28504 keyed
>>28504 I kneel
where the hell do i post something that's not even /e/?
>>28507 here
>>28510 cute
>>28510 I'd consider sideboob to be /e/
>>28512 eeeeeh, kinda? /e/ starts at areola slips imho
>>28513 Ecchi doesn't have to be nsfw pantyshots, underboobs, midriffs and stuff like that all belong to /e/
>>28514 >midriffs are /e/ in what universe
I have a dataset for this artist if anyone wants to take a crack at a vpred lora, i've noticed the artist is sort of fried or inaccurate https://danbooru.donmai.us/posts?tags=dismassd It shouldn't be as bad as the onono imoko dataset since I seemed to have been using a really old tagger for that one (sorry again), I'm now using wd-swinv2-tagger-v3 but I can use wd-eva02-large-tagger-v3 if that's better. https://files.catbox.moe/o8cr4y.rar
>>28515 uohh belly erotic
(3.04 MB 1680x2240 00035-551822852.png)

should I sell my 4090 and get a 5090? I usually buy a gpu and then run it into the ground before getting a new one, but I don't want to get into the cycle of paying $3000 every 5-7 years for one component. My biggest concern is getting trolled by leather jacket man where the performance is actually worse. >>28515 I find the most erotic part of a woman is the tummy
(2.09 MB 1344x1728 00083-2943945638.png)

(1.39 MB 1152x1632 00082-684618946.png)

>>28521 I find that it works somewhat okay mixed into other styles at a lower weight but by itself it's really hard to have it accurately portray the style (whether it be saturation or textured inaccuracies). I'm also using cyberfix, and rescale doesn't help as much as it does for other styles.
>>28522 doesn't look like a worthy upgrade to me but if you don't count your money then why not, it's better
>>28522 wait for benchmarks
>>28526 not yet but i might unironically start grifting soon or my car taxes will bankrupt me
>>28527 Same anon same , got any workflow, ideas or anything that can help grifting ? Im thinking of making clothed versions of some gens and releasing them on pixiv,then teasing censored nude versions so people consider subbing to patreon.
>>28523 Done https://files.catbox.moe/4er349.safetensors I don't know the artist well enough to tell if the bake is good, you have to test yourself
>>28530 thanks anon, I really appreciate it. I'll give it a try as soon as I make it back home today.
(1.43 MB 1152x1632 00043-2151880312.png)

(1.33 MB 1152x1632 00048-192990040.png)

(1.56 MB 1152x1632 00055-549398287.png)

>>28530 it works pretty well, and I find that using things like the style mix in these >>28521 (but with dismassd at a lower weight) give images similar to the softer style of his, which is nice. It seems to work exceptionally well with Coco but that's a given since she is the majority of who he draws.
>>28186 catbox for first image?
(44.76 KB 2202x1492 catbox_t1f0f2.png)

(2.20 MB 1824x1248 catbox_8nkn7b.png)

(2.13 MB 1728x1344 catbox_sg7cxn.png)

(1.91 MB 1344x1728 catbox_h0gv53.png)

>>28522 NAIv4 is flux based, It's only fair to assume local will follow suit once again, if so 50 series will have effectively twice the performance and memory to their 40 series counter part because of fp4 support
>>28536 >NAIv4 is flux based It's a custom architecture, we're using FLUX's VAE but that's it.
>>28526 I want to make an AI patreon or an AI something for follows but all I like or post is loli and i'm pretty sure every site under the sun bans it.
>>28538 me too. >i'm pretty sure every site under the sun bans it. i still post my gens to civitai. their mods leave up a lot of my loli gens, but I won't be surprised if they're gone tomorrow.
>Download lora that claims to do something >Use it for a bit and it seems good >Disable it and the image actually improves I've been seeing this quite a bit for illustrious. Artist loras are great but concept loras are often worse than base.
>>28536 You wouldn't like to use it with nf4. 5000 series will just have hardware acceleration for this precision, quants will probably still run on pure bf16/fp16 performance with upcasting, they should be better and closer to bf16 precision. Presentation is just a bullshit, they comparing uncomparable things, yet to see real benchmarks, probably wouldn't be so much of a gap between 4090 and 5090 as it was between 3090
>>28532 Nice to hear that, gayarou is my go-to softener for non cunny artists although it has non-negligible impact on face structure >>28537 Lol good one >>28543 If LLMs are anything to go by, 10B model in fp4 should be able to consistently outperform 2.5B (SDXL) model in fp16, the sad truth is local in foreseeable future (2 years in terms AI) will be all about what you can fit in 12/16 GB of vram, thanks to leather jacket man
>>28545 I was shitposting but why would they lie about using the entirety of FLUX? They've been pretty open about the base models in the past.
kek i forgot to remove the name
>>28545 haha bunny printer go brrrrr
10 steps is as low as I can go. >>28480 gimp. i can still see the metadata with the script though.
>>28551 >first ever weird, i remember 4bit quants of flux popping up about two hours after the model was released even about utterly trivial shit nvidia can't stop lying
>>28551 >small details completely destroyed can't wait for fp1
>>28554 why does this fucking browser reset shit on 8ch whenever it crashes
>>28557 >knee pits nowhere near as erotic as armpits but sure
>>28560 >didn't catch the reverse legs ok fuck it I'm going to bed.
>>28561 it's okay, she's a wizard
>>28545 It's not even about llms, it's about basic ai scaling laws, jump from 1.5 (870m vs 2.6b) was incredibly huge, that even horse cock was not a problem for a while >>28551 4bit xl lol, the main point to compare precision preservation ig, did they ever release this quasar quantization?
>>28547 3 things combined 1. at some point in 2024 following their successful launch of NAIv3, NAI pivoted towards image gen service from AI storytelling, as a result for the first time their front page is putting more emphasis on image gen then: https://web.archive.org/web/20220605193110/https://novelai.net/ https://web.archive.org/web/20221208033518/https://novelai.net/ https://web.archive.org/web/20230605000844/https://novelai.net/ https://web.archive.org/web/20231205053859/https://novelai.net/ https://web.archive.org/web/20240603031951/https://novelai.net/ https://web.archive.org/web/20241202101646/https://novelai.net/ now: https://web.archive.org/web/20250104040531/https://novelai.net/ 2. image gen has a much lower barrier to entry than LLMs in terms of training and inferencing, NAI is facing hash competition in this area, it's been 2 months since the leak of Illustrious and we already have functioning vpred model built on top of it 3. extraordinary claims require extraordinary evidence, all NAI has ever done so far is finetuning, the only thing (VAE) they disclosed is nothing original That's why I believe these days they are more incentivized to lie or at least be less open about what they are doing
>>28567 >nai is facing harsh competition [...] and we already have functioning vpred model built on top of illustrious what competition? the illustrious' dev/company - tensorart partnership that was announced today/yesterday/whatever and will absolutely be censored? surely you're not talking about local and the botched vpred bake of noob >[...] they are more incentivized to lie or at least be less open about what they are doing not really following your logic here fam, point 1 and 2 are completely irrelevant and point 3 can be paraphrased as "well they didn't make the vae and i don't think they're capable of doing anything other than finetuning existing models because they don't open source their research" and i really don't see how you can follow that up with "well clearly they're incentivized to lie" without sounding mad about them not disclosing something closed source not saying v4 CAN'T be a flux schnell finetune in an oversized trenchcoat after a fp8/q8/whatever crash diet but it's extremely unlikely given how they said that while inference does cost more it's still comparable to XL there sure as fuck isn't any "1-4 steps are all you need" schnell fuckery going on here either
>>28568 v3 has been getting steadily replaced by noob more and more at places i've been checking out. even tech illiterates at boards like trash started switching from from nai because noob is a fully adequate replacement and doesn't cost money and is actually better at sex and degeneracy thanks to e621 shit. sure v3 gives considerably more aesthetic pics out of the box and isn't as jank, but noob is still workable if you are willing to tinker with it (or just use eps), it's free and you can do stuff with it that loraless nai couldn't. and frankly most people don't care that much about their pics being clean, most of nai gens are fucking shit. to me v4 didn't seem like an unconditional upgrade like v3 was. i'm waiting for the full release though then we'll see i guess
>>28569 >thanks to e621 shit. stopped reading there
>>28570 why didn't you rope yet considering your favorite corporation trained a tagger on e621 and most likely is gonna include e621 in the full model?
>>28571 You should go back to the "main” thread where you belong.
>>28572 this thread isn't supposed to be your personal kurumuz fanpage either, but yeah have fun man
>>28573 >v4 is flux >not really it's just the vae >no they're lying it has to be flux >unlikely >well you're a kurumuz-glazing naishill Are you perchance one of those unironic "V3 is a sd1.5 comfy workflow that applies a billion loras on-demand" schizo?
>>28574 my mistake for not making this clear, but my only posts are >>28569 >>28571 >>28573 i actually have nothing against nai and don't think it's based on flux, it's your fanboyism that got on my nerves a bit.
>>28575 What fanboyism?
>>28568 >surely you're not talking about local and the botched vpred bake of noob This is your brain on SaaS
>>28579 >This is your brain on SaaS that's very funny anon, now tell me about v29
Has anyone made a Yunli lora for illustrious? For some reason there's like a billion of them for pony and none for illu. Hope to see one for Tribbie too.
(2.67 MB 1280x1920 00005-3154297863.png)

can the nai vs local fags go back to halfchan, for fucks sake
>>28590 first impressions are that it is much more aesthetic and stable out of the box than noob but it's definitely underbaked
>>28590 What's the official quality tags for it?
>>28593 Should be the same as before
there are some things i like about noob vpred models but it seems like i've got to fight both the contrast and the weird artifacting, ok i've figured out the contrast issue but what do i do with the subtle or sometimes not even that subtle jpeg artifacts? do they plan on fixing both of these things on base level or it's gonna stay that way?
>>28443 that surprised sex face is cute as hell >>28520 witches and brooms... hnng. the toes are cursed tho. >>28583 elves man. hot
bros if anyone can be bothered please bake a lora of this artist https://danbooru.donmai.us/artists/91361
>>28598 what base model? I could take a crack at it, just be warned that painterly styles train unpredictably on SDXL. >>28599 20 images could potentially work but there's 50+ counting their pixiv which is plenty.
>>28600 thanks for deciding to give it a shot >model noob eps preferably, well whichever version out of those is best i've no idea, i'm just mainly using eps versions and mixes
(2.64 MB 1824x1248 01579-3654601562-babushka.png)

(5.02 MB 1536x2688 gr.png)

(5.13 MB 1664x2432 hj.png)

(5.55 MB 1664x2432 pb.png)

(6.34 MB 1664x2432 sp.png)

>>28600 a nice anon in the other place did this lora (or maybe it was you), but if you haven’t started yet or weren’t feeling up to it, you don’t have to. however, if you did start or made an attempt that you weren’t happy with, feel free to share that. either way, i'm grateful that you responded
>>28603 whoa these looks nice, do you have a link to the lora?
is there a mimukauwa nice try miku lora for noob vpred?
I really miss Elins from Tera, and it's a shame that Mojarin doesn't work that well with Noob.
>>28605 nvm there is one but it's a bit janky. might train my own. >>28603 ah I was the one who said I would look at it, I started prepping a dataset but didn't get around to it. was too busy playing video games all weekend. oops. did the 4ch anon make it for eps or vpred?
(5.82 MB 1664x2432 00026-179768352.png)

(5.65 MB 1664x2432 00036-833955683.png)

(4.65 MB 2048x2048 00010-3856986373.png)

>>28604 those weren't done with that lora https://files.catbox.moe/zhu23n.png here's the gestart lora https://www.mediafire.com/folder/pa5cfv0u2t0fz/gestart and fizrot lora for vpred is somewhere in this or previous thread i don't remember the lora i mentioned is here with a preview https://mega.nz/folder/njQmnDQZ#EoKYxBHc29wIMKIWO8Zl1A/folder/HyJSlTyY >>28607 he made it for eps, if you were still gonna bake or if you were gonna do one for vpred that's cool too also that miku is too cute
(1.91 MB 4864x2496 catbox_aoyi67.jpg)

This is not how I expected variables in dynamic prompts work
I love big butts, and big butts on lolis.
>>28611 Doubly BASED.
that's it i'm ditching this shit browser
what model is everyone using? I've had decent luck with Jinkiesnoob
>>28614 https://civitai.com/models/827184/wai-nsfw-illustrious-sdxl I like this one I prompt gacha a lot when bored so it helps that this model has a "stable" style
>>28616 i can promise you the kurumuz shitpost name wasn't intentional that time my mobile browser (via aka glorified webview) resets the name and post body every time i load 8chan
>>28617 I know, I've had it happen to me. I used a name once and browser remembers that for all eternity.
Does local have anything similar like nai's director tools colorize? Just saw that you can use it for making automatic cum variants by just prompting "cum" and the results looked like straight up enabling a layer in photoshop as it didn't affect the original image much or at all, and it even fit the style of the image. Also does anyone have any good tips for enhancing/refining smaller details after hires fix? Normally I do inpaint only masked at 1024x1024 with various degrees of denoise and padding depending on the type of the detail or its location, but that can get somewhat tedious and I wonder if there is a good trick to do it at once or simplify the process. I tried doing a second upscale before, but that usually turns shit as it affects the entire image. Shame there isn't an option that would take a mask and dynamically change the denoise based on the color.
>>28621 I've seen someone on cuckchan mention segment anything comfy workflows, but no results were ever posted so yeah if anyone knows anything it'd be really cool, it's a pretty fun and useful tool.
(540.42 KB 1282x985 1679286963171675.png)

It's been a while now since grifters introduced "adversarial noise", hoping to cash in on angry artist panic. I've seen some bad glazed images here and there, (mostly on shit that'd be excluded from any training data for quality lol) but recently I've seen some more subtle implementation that's almost like a film grain. I'm assuming it's just cope, and even if it isn't that a simple denoise would nullify it. Does anyone know to what extent it actually works, and what current workarounds are? Also, any other AML methods that we might soon encounter aside from like robots.txt and metatag shit?
>>28624 most of the ones i've seen will stamp on some fancy premade box or circle with DO NOT REPOST DO NOT TRAIN DO NOT STEAL DO NOT REUPLOAD DO NOT REDEEM in english and japanese AND whatever shitty pattern they come up with usually at like <50% opacity most jp artists i follow are doing this isn't the pic i wanted to use as an example but close enough
>>28625 are doing it to varying degrees but they're ALL doing something**
>>28625 also holy shit not the pic i wanted to use but now that i zoomed in i'm fucking appalled, that's not even film grain that's fucking 16-bit jpg noise sprinkled on as if it were pecorino cheese on a carbonara
>>28627 also not random noise, i can clearly see some numbers
i'm kind of lost, which model is your go to? vpred?
>>28629 >vpred? Yes. Hard to go back to eps for me. Been using v29 and later switched to v29+1.0 mix that was posted on 4ch.
>>28624 >I've seen some more subtle implementation that's almost like a film grain. That reminds me, the other day I was looking at an artist I was thinking of making a dataset of, and I started wondering if some images had film grain or a very subtle use of adversarial noise.
>>28633 people keep saying things like this and then never produce results because they're wrong
>>28633 have you ever actually used it
>>28631 >Been using v29 and later switched to v29+1.0 mix that was posted on 4ch. link? does it have less noise issues than pure 1.0?
>>28636 can't find the magnet link but yeah https://huggingface.co/autismanon/sdxl_modeldump/resolve/main/NoobAI-XL-Vpred-v1.0%2Bv29b-v2.safetensors hope I won't get killed by chinese. >does it have less noise issues than pure 1.0? I haven't used pure 1.0 much, mostly 0.75 and v29. but honestly I haven't encountered the splotched noise much after vpred0.6 and switching to euler cfg++ simple or just euler simple. this merge dealt with overblown contrast issues that were present even in v29, it tends to be a bit muted color if anything, but I find it easier to work with.
>>28637 sankyuu, though I did find the magnet (bWFnbmV0Oj94dD11cm46YnRpaDoxYThlODBlYjVmYzJlMWRkNDJhZDdmNjhlMTNkMWZlNzNiOWQ4ODUzJmRuPU5vb2JBSS1YTC1WcHJlZC12MS4wJTJCdjI5Yi12Mi5zYWZldGVuc29ycw==) downloading right now and then I'm gonna experiment with it and itercomp merging
(2.86 MB 1248x1824 00070-2902950717.png)

>>28635 i know right? the novelty A.I is just $25. i pay for two, no, THREE NAI subs last month. my cat loo, my wife, and my wife's boyfriend now generate as Krishna intended.
go back to halfchan
>>28639 I enjoy side-tie bikinis, because like me they only require one good tug
how did this pest even get here
take a few more
ah, it's that shitposter
>>28646 I'll spoonfeed you whatever you need to know if you stop acting like a sperg here and leave shitposting to cuckchan
>>28637 thanks for sharing, this one seems to replicate lora styles incredibly well, all kinds of loras seem to work great on it too
v29? v29 of what? First time hearing about such a model.
>>28652 it's one of the test epochs of noob vpred which only some people have access to because for whatever reason LAX never public'd the repo. I gave the merge a try and it feels washed, so I just went back to 1.0 + itercomp.
>>28653 Ah ok, Thanks.
damn i miss momoura now
(2.26 MB 1248x1824 catbox_c5vabc.png)

(1.10 MB 896x1152 catbox_x89dq8.png)

(2.36 MB 1344x1728 catbox_cqy4ws.png)

(2.32 MB 1344x1728 catbox_n9r8bv.png)

>>28656 what makes you assume I'm dead
99% of the time I don't care but some type of posters make me irrationally disgusted and annoyed
i've done nothing to you. all i do is post nyo ,Sora, some stupid comment about nai to make you laugh, and you just lay into me.
you spam ugly nonsensical slop and your "jokes" are out of place and unfunny
>>28664 hnnngggg white pantyhose nopan shondo sexo...
I finally got enough money to buy Nai, I probably cant buy it any time soon again. Can Anons please recommend me some good artist mixes ? I like to use different styles but I'm too retarded to come up with mixes...
Civitai feels fucked for me these days I try to browse it for loras but I get served like 5+ furry bara artists every time even though I have all those tags and like 100+ artists blacklisted
(2.76 MB 1392x1984 catbox_d5l7ir.png)

(3.04 MB 1552x2208 catbox_guqlt8.png)

(2.73 MB 1392x1984 catbox_xmn6ji.png)

(3.15 MB 1392x2160 catbox_zdiivi.png)

(multiple views, comic) coherence is slowly getting better. There are little details that could be fixed with inpainting, but it's pretty consistent with just prompting. (against glass) is a crapshoot but hot when it works.
>>28677 are you not satisfied with noob ipadapter?
>>28679 it felt like a waste of time compared to tinkering with the prompt more, while vibe transfer made prompting feel almost secondary
>>28677 What vibe pics did you even use? Seems like I was underutilizing it when I had NAI. I did use it sometimes but still relied on artist tags more often
(775.78 KB 1108x1584 1452561559654-1.jpg)

(85.39 KB 880x585 1501259338.jpg)

(435.50 KB 2159x1526 1502239913.jpg)

(450.24 KB 850x1235 1507837107525.jpg)

>>28681 anything i thought was nice. i underutilized it too, only started getting into it my last few weeks using NAI. i've seen some mad scientists stacking them six or eight deep to pretty nice effect.
>>28672 >>28673 >>28675 >>28676 Thank you Anons <3 >>28674 >>28677 Idk man probably skill issue but I cant get good results with it >>28678 Yeah I should really try, the money some of these grifters make in a month is more than my yearly income... Any tips and tricks that can help? Any way to automate some of the works, Like maybe wildcards? And the most important part, what styles(mixes) people tend to like more?
>>28683 anytime man make sure you're matching the sampler, some styles are sensitive to that
>>28683 oh, and make sure this shit's getting turned off properly, i remember that was an occasional problem
>v4 drought >nms update thanks sean
>>28686 Ok Anon thanks for help again. My last question before going to half chan and ask there too. any idea what mixes these pixiv geners use? tried to group similar looking ones together, sorry if there are too many, I had these bookmarked long before buying Nai. chiisanaimouto / 110103371 | Blusai / 107747271 | yoyu / 10402649 | nyalia / 8786609 | 天奈R / 109989392 | Satyr_ / 26522981 | mokemoke / 108514503 | 这也是QAQ / 99130349 lativi / 87251195 | サボテンマン / 90418429 | ピピだ! / 103351039 | つるぺた研 / 100729329 | SURUFUMI AI / 34645508 | kanabuns / 103101037 | Half-Ai Creator / 86929043
>>28689 I'm not going to bother looking for you but there's a bunch of the common grifter styles listed here: https://docs.qq.com/sheet/DZWZMemxNZkpVR0VB
>>28690 how the fuck do you save this as a pdf
>>28690 Oh nice, Are there any others like this? I wanted to do similar thing cuz I have a lot of grifters bookmarked with interesting styles
so why are we still calling them grifters? just self-hating totally-self-aware zoomie things?
>>28693 because grifting is the only actual skill involved in making money from ai pictures
>>28694 lolmao ok then lil bwo
>>28690 now to add all of those mixes to a artist wildcard...
(3.37 MB 1392x2160 catbox_3xg6ab.png)

(3.25 MB 1392x2160 catbox_r6zq13.png)

(3.32 MB 1392x2160 catbox_pztc09.png)

(3.19 MB 1392x2160 catbox_ck0atf.png)

more multiple views slop. The anatomy is definitely off in little ways. Looong bodies and all that. But still good enough imo, when you want clothed and not-clothed, without having to mess with something like prompt grids.
>>28693 it's a lazy loan word i substitute for hustle. >>28696 you're a fucking braniac. here i was manually copying each gen to a folder by mix like a 'tard.
>>28698 if you want to be real high IQ like myself, I held shift + alt and dragged from top to bottom, copy paste into notepad ++ and then ctrl + f all of the {{ or [[ into (( ))
(1.24 MB 832x1216 00109-1297623428.png)

Is there any solution for doing automatic latent upscale on reforge's hires fix and keeping the APG configuration? >did you compare hires fix and img2img latent to make sure the config isn't being reused by default Yessir. CFGs of like 8 or 9 completely burn images on the hires fix because the APG isn't doing its thing.
(784.50 KB 768x1024 00159-4098043611.png)

(1.16 MB 768x1024 00153-4098043605.png)

(1.18 MB 768x1024 00141-104781652.png)


>>28699 i used vim. that second part about manually changing each "{{" sounds like work though. there's probably a lazier way.
>>28693 Grift is hobby + monetization. Hobbies not only survive but thrive without it.
>>28701 thanks bro
(1.07 MB 896x1152 00014-2321969028.png)

Is gay shit fine?
(3.17 MB 1344x1728 00242-2486571257.png)

>>28706 Okay, got it.
>>28705 >>28707 i want to see. post it here: >>>/h/5336 i think you'd have to spoiler it tho.
>>28709 aronabox? curious what you're doing here with the colors
>>28712 Thanks for reminding me I need to watch the latest part
>>28713 Every time a new part comes out I'm tempted to rewatch the whole series. Maybe I'll just read the chihatan doujin
>>28715 I sometimes rewatch the Anzio OVA just because it's the only substantial bit of Anzio content we'll ever get. Fucked up how that OVA is 10 years old and the first season is like 12, how are we still not done yet?
(1.88 MB 1600x1200 catbox_dppdlr.png)

(1.80 MB 1600x1200 catbox_19ldzq.png)

(1.82 MB 1600x1200 catbox_c2dadg.png)

(1.86 MB 1600x1200 catbox_bjgvvf.png)

>>28717 what on earth do they need three nintendo switches for
>>28718 When Li-kun comes over.
Greetings Anons. Any recommend tools to use for fast censoring pictures and metadata removing (in large scale) ?
>>28722 stupid... stupid sexy trial captain...
>what's wrong with my inpainting? everything keeps getting darker every pass >is it the artist mix? is it the inpainting model? two days of that before i realized that i had vectorscope on
>>28724 base gen box nyao (please)
(1.96 MB 1600x1200 catbox_zshg8q.png)

(1.89 MB 1600x1200 catbox_p8rjjg.png)

>>28726 what the fuck how long has stealth-pnginfo been disabled on my install
>>28697 it looks clean. >>28710 another thing, lowering the weight of partially_colored also makes it look really nice.
>>28731 about time that zigger delivered, EPS is useless for me because noob 0.5 still mogs if you want something stable. will definitely crack this open and gen some cunny later.
>>28731 so far i don't care for it
(3.14 MB 2464x9535 catbox_89aywh.jpg)

>>28738 Great concept 6 fingers in the second pic but still great
>>28731 Neat.
(2.00 MB 1344x1728 00038-3416296213.png)

Is it just me or does Rouwei do the thing where it usually produces completely unhinged backgrounds if you don't give it some guidance on what you want?
(2.45 MB 1344x1728 00059-3109928601.png)

I do like the way it does styles, though.
>>28738 really cool. it could work as a 4koma too.
(86.40 KB naiv4athome.txt)

(1.68 MB 1152x896 00006-1056485597.png)

(1.35 MB 1152x896 00007-2926627291.png)

(1.25 MB 1152x896 00008-1723006080.png)

>>28701 >>28704 i removed the brackets and curly braces.
Have english vocaloid fans always been insufferable or is it just kneejerk moralfags from tiktok
>>28741 box for this one?
(1.97 MB 1200x1600 catbox_y5prsm.png)

(922.45 KB 896x1152 00008-2169169300.png)

>>28747 https://files.catbox.moe/fclnoh.png Upscaled from 896x1152 by 1.5 times. I'd recommend the settings in https://files.catbox.moe/am4fcq.png for rouwei 0.7 though.
>>28746 >NoobAI-XL-Vpred-v1.0-cyber4-iter+v29_03 Damn...
>>28746 >>28751 Also curious about this, did you make it yourself with the cyberfix fomula? Base v1.0+v29 is too broken for me but I like it's style
>>28746 >Have english vocaloid fans always been insufferable or is it just kneejerk moralfags from tiktok it's young libshitty moralwokefags, it's always been this way but it's been getting worse post-2010s
>>28750 this is exactly my fetish wth
>>28754 >Base v1.0+v29 is too broken for me In what way? I've been pretty happy using it, although I prompt artists and use loras all the time, I never just use the bare model. Cyberfix is admittedly easier to use but it kills styles way too hard which is pretty much a deal-breaker for me.
>>28758 ooh, can i get that muyu lora?
>>28749 honestly my gens on this model come out extremely mangled, no idea why though
(2.83 MB 1344x1728 00001-1329589022.png)

>>28761 >enable APG/guidance limiter >bump CFG up to 7 or 8 >SGM Uniform and DPM++3m sde or 2m, around 14 steps >remember to add some tags in the end for the background That's all I'm doing. >isn't CFG 7/8 too high for vpred? Sort of? I skimmed through the APG paper and it's supposed to allow you to crank CFG way up. Some models actually still produce better results with like CFG 2 and APG for some reason, but Rouwei seems to need it.
(4.18 MB 5544x10712 catbox_bwjzwv.jpg)

>>28758 Thanks, found a ready-made mix on huggingface https://huggingface.co/Toc/toc/blob/main/models/NoobAI-XL-Vpred-v1.0%2Bv29b-v2-perpendicular-cyberfixv2.safetensors >>28757 This >>28739 for example Is APG and the like mandatory for v29? Have been using CFG++ samplers for too long
(8.85 MB 4864x2496 file.png)

(14.51 MB 3328x3648 file.png)

(9.92 MB 4032x3648 file.png)

The last few /hdg/s were, dare I say, good
Is there even a single usable AI thread on halfchan? /hdg/ is.. well, /hdg/, /edg/ is practically dead, /slop/ has actual posters but has literal discord faggot erping hours and has a meltdown the moment you point it out, /vtai/ is actually nice minus the kiaras and moris but it's too limiting.
>>28769 If you don't mind potentially missing the rare productive conversation you can press F and suddenly threads becomes much more bearable. /slop/ and /vtai/ are probably the most usable you mentioned. The /jp/ general is decent too. Nothing is as good as it used to be though.
(60.17 KB 1747x830 Untitled-1.png)

Is the AI threatening me?
>>28769 I'm mostly lurking lately but the /jp/ thread seems to be the best one right now. But it's *only* 2hus.
>>28773 > But it's *only* 2hus. Eh, not *that* big of a problem honestly.
>>28772 what was the pic?
is civitai just unusable for looking up loras now? every time I try to browse all loras then I get jumpscared by gay bara shit or some furry even though I have like 200 of these fucks blacklisted along with every tag they use
(1.82 MB 1283x883 Capture.PNG)

>>28776 I kept scrolling down and the first bara one was the only one that showed up, so I assume it's fine.
>Illustrious 1.0 and 1.1 are finally out >Both paywalled
How do anons keep up with this hobby... Too many models to download Too many boards to follow Too many good gens to save Too many "oh this style that this anon posted looks good I should try it myself too" I have to spend hours daily just to keep myself updated, I rarely even have time to do actual generating
what the fuck is going on on halfchan
>>28783 >starts with m ends in e. if you're so bored that you need to shit up two generals on two different sites you should consider making yourself useful and donating your blood and organs right now fam.
>>28784 he loves the negative attention, just try to ignore him. it almost feels like he's underage tbh
>>28785 it's kinda funny seeing people practically ignore his literal shitposting but me ban evading two or three times per thread just to keep posting cunny triggered literal nuclear meltdowns despite me not really saying anything every time
>>28786 the times are different, there was much more drama on hdg. also he seems to be really dumb, it's like getting angry at a misbehaving animal im not sure that it's actually him flooding thread with shit though, his confession looks like a shitpost.
(1.81 MB 1344x1920 1738847725261465.png)

(12.15 MB 3328x3648 file.png)

>>28775 This one from the other /hdg/ Tried to replicate the out stretched arms pose but couldn't make it stable
>>28789 they behave the same way and he regularly shitposts there too, wouldn't be too far out there
(2.08 MB 1344x1728 00009-4099643888.png)

>>28779 I'm going to cry. >>28790 >Tried to replicate the out stretched arms pose but couldn't make it stable I'm impressed you got it to work at all, I've been trying for fifteen minutes and this is my only success (scuffed hands).
>>28792 hi momoura illustrious 1.0 is pay to download on tensorart and (allegedly) already leaked https://boards.4chan.org/h/thread/8445434#p8448443 it might be dolphin porn or a renamed model and my replacement 3090 is still in its box so uh someone take one for the team
good thing i have drawthings set up on my ipad, huh, faggot :)
(2.19 MB 1344x1728 00026-1530176165.png)

>>28793 >hi momoura Not him, just borrowing the possummachine artstyle. >it might be dolphin porn or a renamed model and my replacement 3090 is still in its box so uh someone take one for the team Can't we j— >they hid the model hash on civitai I think I'll just pay 10 bucks, thanks for the spoonfeed.
how do you even collect data for a lora if shit isn't tagged properly on boorus like https://danbooru.donmai.us/posts/8815769 there probably isn't even a tag for this
Are there any other place for cunny gens than here?
>>28798 /degen/ on /b/ but the average gen is much worse than /slop/ (the latter allows clothed cunny though if you care)
>>28796 it seems like it's actually legit but yeah, best if you pay up yourself or just wait a day or two, someone else will probably leak it apparently the release cycle is exclusive on-site > new model > pay to download > new model > free to download so the free one is always 2 versions behind and the paid one is just 1 given how quickly 1.1 came out you'll probably be able to pay for it soon enough
>>28552 Ik I'm a little too late but any chance I receive box? Love the style, seen it many times but don't think I have the prompts/models for it
>>28796 (Me) >they only accept fucking paypal Never mind then. >>28800 Neat, good to know.
>>28797 Just search for armpit_focus and keep scrolling while looking for images.
(6.57 MB 2496x3648 file.png)

>>28792 It's really hard to get right, and the whole thing crumbles down the moment I add any artist or character to it
>>28803 anon that'll just lead to me jerking off i tried presenting armpit + bodysuit and it seems like that's literally the only pic of a bodysuit with an armpit zipper i'd ironically curse the artist for giving me a new fetish with zero material were this <2022 but with enough hours of inpainting and scribbling i could probably approximate it though i wonder if v4's nlp or even vibe transfer would help in any way, probably not in the case of VT
>>28796 i mentioned momoura because he's the only other KF/kaban poster on here
>>28790 Been looking for decent wildcards to use, if you don't mind plz share it :3
>>28808 Kaban is the first character in the grid in >>28790 and that's why I used her.
>>28810 >there's still only 2 of us we're going extinct...
>>28796 >>they hid the model hash on civitai They did not. You can check the metadata of the images posted, and then use the API to verify it. https://civitai.com/api/v1/model-versions/by-hash/735cf3fefc
>>28812 Alternatively you can just call the endpoint for the model version. https://civitai.com/api/v1/model-versions/1389133
>>28813 Can confirm, the leak is legit [I] Stable-diffusion> sha256sum Illustrious-XL-v1.0.safetensors 735cf3fefcbdc4f7817f53247e38b836ffd27c7641af6d8daa21d245242cb4bd Illustrious-XL-v1.0.safetensors
>>28793 The link is down already, anyone got a torrent or a mirror?
>>28815 supposedly the same hash http://pixeldrain.com/u/Q6PqNTUN again, haven't verified it myself
>>28812 >>28813 Thanks for the correction. I noticed the hash was missing from the UI and assumed the worst (incorrectly).
>>28816 and also a torrent, ditto PBIE4L73
24 in a row are probably enough
>>28770 >>28773 /jp/ thread is pretty slow but does have plenty of discussion, looks pretty comfy, thanks for pointing out the thread even exists
>>28768 whoa can you share a catbox of one of the third ones? with ame/kangel
>>28830 nta but if I had to guess they took it from >>28758
(1.31 MB 896x1152 00291-2341152123.png)

(1.02 MB 896x1152 00284-2986006047.png)

(2.06 MB 1344x1728 00034-4063396472.png)

(1.36 MB 896x1152 00295-1760332363.png)

Illustrious 1.0 definitely exists and produces coherent outputs some of the time. I tried out the native 1.5k by 1.5k res but that seemed to just worsen things, does anyone have good settings I can just copy? Using APG + SGM Uniform + DPM++3M SDE 16 steps.
>>28833 Nvm I forgot about the epred cfg thing.
>>28833 >Illustrious 1.0 definitely exists and produces coherent outputs some of the time. Basically my thoughts on the model. I'll go back to my 0.1 finetunes that work instead while I wait for the 1.0 tunes.
(3.18 MB 3696x11172 catbox_30w9y0.jpg)

>>28833 It's undertrained by a mile compared to latest noob, that's for sure
>>28836 Where can I get this v29b-v2... ?
(8.64 MB 13440x3928 xyz_grid-0000-2554261280.jpeg)

CONVERGE ALREADY DAMN IT STOP TURNING HER FACE AROUND 50 STEPS IN >>28835 >while I wait for the 1.0 tunes. Can there be 1.0 tunes? I'd imagine there are licensing issues since you literally need to buy the weigths.
(11.61 KB 1062x224 Capture.PNG)

>>28838 >Can there be 1.0 tunes? I'd imagine there are licensing issues since you literally need to buy the weigths. Nevermind, yes.
>>28839 >sell this model or merges are they actually retarded? what's stopping me from renaming the model and selling it?
>>28840 I could have screenshotted the wrong thing? Might be a good idea to double check I'm not misreading the page. https://tensor.art/models/828830887698374324/Illustrious-XL-v1.0-v1.0 >what's stopping me from renaming the model and selling it? I don't think you'd even need to rename it?
>>28841 wow. the gooks are actually retarded, huh.
(10.79 MB 4800x7764 ka1.jpg)

(11.37 MB 8064x5888 ka2.jpg)

>>28831 >>28832 it appears i am double retarded sankyuu
>>28837 I downloaded it from a random huggingface repo, I don't really know what it is >>28763
>>28839 >>28840 1.1 has the same perms by the way so you can make a merge with 1.1 or make a lora using 1.1 and just put it out
What's considered the best sampler? I've been using Euler Ancestral CFG++
>he has a """personality"""
>>28848 I use euler a / sgm uniform
>>28850 scheduler seems more impactful than sampler to me, honestly
>>28851 yeah it seems like Beta and SGM Uniform are the agreed upon schedulers
>>28848 I like DPM++3m SDE and Kohaku-LoNyu-Yog. DPM++3m SDE does this thing where, depending on the schedule, if you leave it on for too many steps it just introduces noise into single-color backgrounds like JPEG artifacts, so watch out for that. KLY burns through a lot of computational effort but converges really fucking fast.
>>28853 which schedulers/steps are best for those? I see a lot of people use really low steps for DPM++ SDE
(9.28 MB 2496x3648 grid-0130.png)

>>28855 cute tiger
(1.90 MB 1200x1600 catbox_464ldi.png)

(2.05 MB 1200x1600 catbox_qt6x3r.png)

(4.29 MB 2688x1792 grid-0102.png)

A shame interlocked fingers are broken, otherwise a good concept
(9.39 MB 22176x3834 xyz_grid-0007-2082819307.jpg)

>>28851 Thankfully there are only like 4 different schedulers and then worse versions of those schedulers. >>28854 Picrel looks funny but DPM++3M SDE with AYS usually converges correctly in like ~12 steps.
>>28862 Some more funny grids: https://files.catbox.moe/un0ufl.avif https://files.catbox.moe/xkls2x.avif https://files.catbox.moe/3r1eo0.avif This one is particularly neat if you want to see what's up with Kohaku-LoNyu-Yog: https://files.catbox.moe/x99v4h.avif
>>28863 Did you use the same CFG value for CFG++ and ordinary samplers? There shouldn't be that much of a difference between Euler CFG++ and Euler
>>28865 I want to be more horny but lately I'm just not.
>>28866 ditto
>>28864 It's all CFG 2, APG enabled with the default settings. I don't think I could even use different CFG values on the XYZ?
(6.98 MB 5776x10932 catbox_er6u32.jpg)

>>28868 That explains why your AYS gens are covered in weird textures, they do be like that on low CFG And yes a fully automated fair comparsion between CFG++ and ordinary samplers can only be done in ComfyUI
>>28869 >That explains why your AYS gens are covered in weird textures, they do be like that on low CFG They usually look fine tbdesu. I think I just rolled an unlucky seed. >And yes a fully automated fair comparsion between CFG++ and ordinary samplers can only be done in ComfyUI Neat. I don't have comfy set up right now, though.
(2.22 MB 1600x1600 126802405_p0.png)

(1.91 MB 1400x1400 127203752_p0.png)

(466.44 KB 2048x3072 126928962_p0.jpg)

(416.73 KB 2048x3072 126969463_p0.jpg)

Seeing a lot of this style recently, do Anons know what model or mixes they use? I fucking love the style would appreciate any help lativi / 87251195 サボテンマン / 90418429
>>28874 One of the styles involved is definitely healthyman, seems to be mixed with something that has shinier shading or possible a checkpoint merge with somewhat homogenized style.
>>28875 Thanks for help, I guess I have to ask in half-chan too, hopefully they wont delete my post if I put the cunnies in a box
>>28874 >サボテンマン I believe his mix has wagashi (dagashiya) in it https://www.pixiv.net/artworks/124769610 https://www.pixiv.net/artworks/125148466
>>28874 I find it interesting the NAI user (lativi) has 0 paid material, but the same cannot be said about the local user (サボテンマン) Anyways lativi is definitely using healthyman and tsurui, maybe some others but those are for sure the main contributing factors. Also they seem to intentionally include "sketch" or similar to make it feel more rough. Here's a quick test I did, first pic is from https://www.pixiv.net/en/artworks/126417644 and second is NAI (here's txt2img if you want the base https://files.catbox.moe/s3ljdc.png)
>>28878 Also while testing this I forgot how fucking fast NAI is H100s + (likely) TensorRT is just unmatched compared to anything you can do locally unfortunately.
>>28879 4-5 seconds for a pic is pretty great, yeah though I swear they've reserved enough compute for V4 to make V3 feel noticeably slower than usual
>>28878 >>28877 Thanks for help, I think our only hope is for someone to make a lora based on their works. Is it even possible to achieve good results with censored/sfw as training data?
>>28881 >Is it even possible to achieve good results with censored/sfw as training data? yes
>>28880 >though I swear they've reserved enough compute for V4 to make V3 feel noticeably slower than usual yeah, no, v3 is way, way slower than usual while v4 is lightning fast
>>28881 Yes, see >>28836 That artist never do porn
>>28881 >sfw Absolutely, I'd argue a lot of times it's better than censored dataset or dataset with poorly drawn genitals
https://mega.nz/folder/ctl1FYoK#BYlpywutnH4psbEcxsbLLg Is the anon who owns this mega still here? I've seen some gens made with guweiz lora of yours but I don't think you ever posted the lora itself. Would be very nice if you shared it, thank you
Has anyone finetuned illustrious 1.0/1.1 yet? Anything interesting?
Guys what's currently a better model that doesn't harm innate style? I thought WAI was the best at first but I think it influences artist tags way too much.
>>28888 not yet, illu 1.0 seems to be pretty scuffed from the unusually high training res. it's also maybe marginally cleaner than noob but trained like 10% as much. >>28889 I'm still using noob vpred 1.0. specifically, this shitmix: https://huggingface.co/mirav/TestMixes/blob/main/vpred1.0_00006__fixed.safetensors not sure what the recipe is but it's given me nicer results than any of the itercomp/cyberfix/v29 mixes i've tried
>>28889 WAI is the best if you like to prompt gacha without artist tags If you actually use artist tags though, I'd use literally anything else. Probably Noob, base illu or something else.
>>28896 style works a lot better for hex than the lolis desu
belatedly recognizing that "masterpiece" has an anti-body type c effect
>>28899 Absolutely possible, taggers are just classifiers where you set a confidence threshold. Would you be able to 1:1 extract the exact artist weight ratios? Probably not, but you'd be able to get a damn good idea of what's in an image. The hardest part would be data curation and you could probably set up a pipeline to generate a bunch of synthetic data on NAI/Noob for it.
>>28898 ... what?
>>28899 wd14 taggers could be able to do this if artist tags were not removed from the dataset
(3.13 MB 2288x9884 file.jpg)

(3.60 MB 3696x9925 file.jpg)

>>28898 Which model? Masterpiece means totally different things on different models
>>28901 only real gamers will get it >>28903 noob, it's usually not a big effect and it's more noticeable if you're using non-loli artists but i went through and tested my artist mixes and about half were improved by dropping it
>>28907 UOHOLYSHIT THAT SHINOBU
>>28907 1girling is good for the soul
>>28907 Agreed, holy based Shinobu
Btw, at some point reforge added support for AVIF, so you can switch to that and get a pretty big reduction in storage usage (at the cost of having to manually convert images to upload them to most places). >does it store metadata in avif correctly? Ye.
So this is why 4chan userscript cannot parse non-ascii metadata in jpeg const exifMeta = String.fromCharCode(...exifReaderMeta.UserComment.value.slice(9).filter((value) => value !== 0));
>>28916 Fuck, I thought I replaced all that with TextDecoder. Updated now.
>>28917 Since you are here, is there any particular reason you didn't drop exif-js? exifreader seems to be working fine on both avif and jpeg files
>>28919 he says it on his page, it's Live2D
>>28918 Because I didn't think to for some reason. Updated again.
>>28915 Thanks for the hard work.
I started pixiv recently. Should I give up on effort posting (sfw) on pixiv? I spend hours inpainting and promoting only for 300 people to see and 20 to like them. Should I just mass generate on wildcards like most of the other grifters ?
>>28927 Could post it in >>912 Also it's fucking crazy looking at that thread now and only seeing one account still active.
(1.44 MB 1152x896 file.png)

(184.80 KB 1152x896 catbox_bpk0b6.jpg)

(184.80 KB 1152x896 catbox_bpk0b6.jpg)

(118.53 KB 1152x896 catbox_j5w37s.webp)

>>28927 as far as i'm aware if you wanna get traction you need to do fotm nsfw >>28928 i stopped posting on mine once naiv3 made it easy, to be honest
>>28928 lmao >>3725 this is really funny to look at after 2 years
>>28931 2 4 with that style looks too much like a real 6 year old's proportions, it's weird.
>>28927 how many followers? >>28928 pixiv graveyard.
>>28931 i have to see their pussies first tbqh
>>28931 #4, kanna proportions
>>28932 damn, this board used to pretty active back then but yeah glad it's comfy if slow now.
(3.68 MB 2096x1664 catbox_ma60yy.png)

(3.64 MB 2096x1664 catbox_d0e0ke.png)

(480.93 KB 1728x1344 00477-623684272.jpg)

>>28938 Whoever came up with the idea of software randomly injecting identifiable data into file metadata is getting sent balls-deep to the depths of thundering and roaring fatal niggerhell, right next to the guy who made it so Word saves by default to your work OneDrive account.
>>28940 I love this goat
(256.69 KB 1152x896 00921-2687908256.jpg)

>>28942 wide_face her for maximum loveability.
>>28943 wide mouth for wide snacks
(227.18 KB 1152x896 00932-2184934491.jpg)

>>28944 Wide snacks for wide mouths.
>>28938 what metadata does webp store? >>28927 >hours inpainting and prompting >hours what's taking you so long?
(162.66 KB 1152x896 yuuma vacuuming chicken leg.jpg)

>>28946 For just 10 dollars, you can buy Yuuma 10 dollars in fried chicken.
(1.57 MB 1824x1248 00025-2148684194.png)

What's so special about v29b cyberfix? I've still gotten decent results. but overall EPS models like personalmerge are just much less of a hassle in heneral
(274.94 KB 1344x1728 00023-3279535960.jpg)

>2girls >OCs >any degree of reliability or consistency across gens This is basically impossible right now, right?
>>28951 yes but maybe regional lora works, now make more of what you posted
(1.50 MB 1536x1536 file.webp)

Test
>>28947 Not even GPS geolocations are removed >>28953
(440.70 KB 1344x1728 showing off panties.webp)

(350.59 KB 1344x1728 kiss 1.webp)

(512.25 KB 1344x1728 kiss 2.webp)

(483.82 KB 1344x1728 fingering through panties.webp)

>>28952 >regional lora To elaborate, I mean OCs as in "my OCs" (as in, "I've combined a few tags and I'm calling this character design"). >now make more of what you posted Me sloppa es su sloppa.
(401.24 KB 1344x1728 hand in panties.webp)

(358.34 KB 1344x1728 fingering 1.webp)

(371.07 KB 1344x1728 clit stim.webp)

(466.96 KB 1344x1728 spread pussy.webp)

(367.15 KB 1344x1728 fingering 2.webp)

(385.15 KB 1344x1728 fingering orgasm.webp)

(352.41 KB 1344x1728 showing off squirt.webp)

(499.22 KB 1344x1728 pussy spread final.webp)

(310.86 KB 1344x1728 00140-1880230407.jpg)

>>28957 Genuinely wondering if it was retarded of me to not leave them both naked, I like the head_between_breasts squish, but I also like "undressing just enough".
>>28955 probably best to bake a lora if you want consistent ocs without having to inpaint every time. you can probably get it SOMEWHAT consistently just with tokens if you autistically prompt every feature but that's a waste of tokens and will probably lead to ocs mixing together if you don't use regional stuff also i didn't know i needed megumu molesting aya until now
(412.07 KB 1344x1728 00160-2964857809.jpg)

>>28959 >probably best to bake a lora if you want consistent ocs without having to inpaint every time. you can probably get it SOMEWHAT consistently just with tokens if you autistically prompt every feature but that's a waste of tokens and will probably lead to ocs mixing together if you don't use regional stuff That sounds like an absurd amount of work either way. I think I'll just do my 2girls scenarios with 2hus. Thanks. >also i didn't know i needed megumu molesting aya until now Pre-dicking lesbian diddling is absurdly lewd. Also, if you have more duo ideas, I'm listening, the only ones that really click for me are Megumu and Aya, Seiga and Yoshika and picrel.
>>28960 >That sounds like an absurd amount of work either way if you want to try the "fuck my tokens" approach you can just make an embedding (the "bash together a bunch of tokens in a text file" kind) for each oc and try your luck, either way you're probably gonna need to inpaint >Pre-dicking lesbian diddling is absurdly lewd agreed though my favorites are the "one girl presenting the (embarrassed/uncomfortable/unsure) other" ones (with or without the lesbian diddling but preferably with, cherry on top and whatnot) seija and shimmy/aya and cirno btw
>>28961 >Aya and Cirno Any specific reasoning for that one? Sounds lewd tho.
>>28962 not too sure tbh, artists started drawing them together with aya as a massive lolicon in the lewd ones and it just clicked
(257.39 KB 1344x1728 00352-1555035616.jpg)

I'm contemplating it. This is less consistent than Yukari and Okina btw, half the gens have Aya on Cirno's lap for whatever reason.
(304.42 KB 1344x1728 00479-2604198191.jpg)

>can't gen clitoral injections without a LoRA It's over. >>28965 Very nice hands. Did you have to do any inpainting?
>>28967 Yes and some image editing first as well.
>>28967 >>can't gen clitoral injections without a LoRA and that's a good thing (the lolis need to be smaller btw)
(174.20 KB 896x1152 00779-3427492687.jpg)

>>28969 >(the lolis need to be smaller btw) The gens were they're tiny tend to have worse compositions, I think. See how she's visibly trying to escape in >>28967 versus picrel. I think it's the style, I'll try some others.
>>28970 shame, it's a good style
>>28971 I think mixing in hamada_yoshikazu and adding the height_difference tag improved things.
>>28973 nice now try seija molesting shimmy
>>28973 Nice. Is it just them and spacezin? I'd be interested to see the params
kk you need to come home the kids keep asking where you are and they're no longer buying the milk card
(471.61 KB 1344x1728 00540-856198738.webp)

(380.55 KB 1344x1728 00569-3113622799.webp)

(429.82 KB 1344x1728 00580-1375613147.webp)

(435.59 KB 1344x1728 00587-2569103181.webp)

>>28975 Here you go. This feels evil. >>28976 It's not spacejin, it's a western artist. The gens attached: https://files.catbox.moe/ep195y.avif https://files.catbox.moe/z9hg75.avif https://files.catbox.moe/vqnaim.avif https://files.catbox.moe/38zezw.avif And the base pre-upscale gen for the last one: https://files.catbox.moe/49bvbx.avif I'm not doing anything special, I think.
>>28978 Ahhh right, them. Thanks for the share though!
>>28978 NICE >This feels evil if evil why hot
(354.59 KB 1344x1728 00599-4085082394.webp)

(379.46 KB 1344x1728 00603-3472407623.webp)

>>28978 (Me) Sorry I didn't notice the boob misalignment earlier, better gen. >>28979 Yeah actual spacezin looks more like the second attached picture for some reason. Maybe because I'm not using the noobai newest tags stuff.
Random question but do you guys use NVIDIA Studio drivers or regular Game-Ready drivers? Does it make a difference?
>>28983 Game Ready is bleeding edge and Studio is stable. I use Studio because I work in lots of other 3DCG software and I've been burnt by crashes and shit flat out not working on Game Ready several times.
>>28984 Thanks. I was wondering because NVIDIA actually lists ou ML-related optimizations for the Studio driver so I wasn't 100% they were ported over between each other.
>>28983 >>28984 nta but there's like a <<1% difference at worst between the two in both average fps and 1% lows in 98% of games, sometimes the studio driver actually performs slightly better (as in outside margin of error better) in a few games but it's like 3090 vs 3090 ti kind of games (<10 fps) the only game where it's strictly worse is f*rtnite so you might as well keep the studio driver and forget about it
>>28986 kind of GAINS* thanks autocorrect
>>28988 why is her head so fucking big
>>28989 it's 0huton
>>28989 big thinkies
(2.95 MB 1840x1696 catbox_mgoayy.png)

(2.91 MB 1696x1840 catbox_n7lg2l.png)

(2.55 MB 1600x1736 catbox_7qwvpa.png)

i wish 4chanx actually worked on here, werk mode would be nice
>>28993 Making a violentmonkey extension that just hides every image with ".media/8" in the URL would be pretty simple, but you'd need to enable/disable the extension directly on the violentmonkey menu.
>>28994 bold of you to assume i remember anything from my high school js classes
>>28995 nta, you don't even need JS for that... it's a CSS rule. img[src*="/.media/8"] { display: none; }
>>28996 the 8 in the >>28994 post was a typo
>>28997 I might be hallucinating but I don't think I even typed that, might be a text filter.
>>28996 oh. i was ready to get ST set up all over again in an attempt to wrangle some llm into wiring that and fixing the booru script but thanks
(11.02 KB 331x339 Capture.PNG)

>>28998 Surely there isn't a text filter that turns ".media" into ".media8"
>>29000 .media/8?
>>29001 Oh it needs the / at the end. Wack.
>>28999 >was ready to get ST set up all over again in an attempt to fix the booru script i still want to do that but uuuuuuuugh i don't feel like visiting and lurking in /aicg/ for whatever breadcrumbs proxy MIGHT currently be active
>>29003 Just put 20 dollars on OpenRouter and ask your questions directly on the website.
>>29004 i'm paying for nai but i'd rather burn my money than pay a single cent for textgen
>>29005 >than pay a single cent for textgen You're already paying for textgen from novelai though.
>>29006 i fucking knew you'd say that yes, fine, have your gotcha
>>29006 >>29007 i tried to use it back when i didn't know about proxies and man it's pretty awful then again, using a text complete model as a chat complete one isn't the smartest of ideas
>>29007 >>29008 You can also use the deepseek app on your phone for free. >is it good for assistant tasks Yeah. But like, 20 dollars for random coding questions will last you months on OpenRouter.
>>29009 i don't code, the last time i touched a single line of code was for the booru script courtesy of i don't even remember which llm and the last time before that was nearly a decade earlier i'll give deepseek a go
>>29010 deepseek's alright, chat feels moderately less slopped than gemini and maybe a little dumber and reasoning can spit out some gold when it isn't eating your entire output thinking main advantage is that it's fucking cheap, like 1/7 the price of sonnet 3.5, and they promise they don't log for all that a chink promise is worth
>>28992 Could you please share the lora. Looks so good!
>>29013 nta, the LoRA is barely doing anything there but it's this, same hash: https://civitai.com/models/954817
>>29014 Gotcha. Thought it was the Lora that is doing the that distinct Amamizore face style. But maybe it's the effect of the tagged murata range and kantoku. Thanks anyways!
(1.17 MB 832x1280 17333-20250223041459.png)

(1.03 MB 832x1280 17245-20250222141750.png)

Made a LORA based on the artist WaxReplica or ShiShigaJ5 who created Monaca (if you know you know), some anatomy issues with elongated torsos and such but I'm satisfied with the result, will revisit a bake possibly. For Illust v0.1 eps https://files.catbox.moe/cltdfw.safetensors
>Prompt hires >AI gives me this
(3.93 MB 1696x1944 catbox_t8ol62.png)

(4.34 MB 1824x1944 catbox_4c7zee.png)

(4.35 MB 1824x1944 catbox_hnwmxe.png)

(3.68 MB 1472x1944 catbox_6gnvb5.png)

(537.24 KB 1344x1728 00447-4221415464.jpeg)

white_theme goes hard for SFW, I think.
>>29021 berry acute style
(459.23 KB 1344x1728 00468-3538483784.jpeg)

>>29022 It's https://danbooru.donmai.us/posts?tags=harano_kaguyama+&z=5 diluted a bit with a softer style. It's interesting because you can tell that the model is copying the style correctly, what fucks it up is the composition. Everything still takes 10 gens for a success tho.
>>29021 >>29023 fishing lora anon is shedding tears of joy...
(1.96 MB 1600x1200 catbox_p9w5qa.png)

(1.84 MB 1600x1200 catbox_y4g95m.png)

(1.92 MB 1600x1200 catbox_rlyysz.png)

ccsakura and tentacles.
>>29027 Do you mind sharing your wildcards ? Im getting tired of gening same things over and over, my brain cant come up with anything new
>>29028 just take a break
>>29029 Dont get me wrong i like and enjoy it, its just i cant come up with anything new
>>29028 Just come up with doujin plots in your head and then gen the scenes in order.
>>29033 Yeah ig that works too still will w8 for Anon to share their wildcard its useful for when im just gonna turn off my brain and do lazy gens
>>29032 I think I remember that style. Was it kuvshinov, askzy, cilo, kemuri haku and something else?
>>29036 Almost, here's a catbox for the base gen. https://files.catbox.moe/9xmpa7.png Been going through my gens and pics to sort everything properly, god it takes forever
>>29037 Seems pretty similar to those gens, params are very identical. Are those yours either?
>>29038 Yes.
(305.15 KB 1728x1344 eiki fishing 3.jpg)

>>29035 NTA but very nice extension. However, I can't seem to be able to use it to gen multiple poses/outfits for a given character because other character names and the like keep leaking in. Is there some trick to it? Am I supposed to regex out hair/eye color tags with the "Tags to Remove" option somehow?
>>29042 >total slop big tiddy reimu wtf i love her
>>29042 Okay, didn't know that, there is something wrong with kohya lora training on vpred, not the thing with uneven layers that some anon mentioned before on 4cuck, but tendency to go back to how epsilon is functioning during train time, especially during longer runs, despite of the settings or base model (there is just 2 but whatever), even just simple --v_param --ztsnr will lead to that for some reason. Alot of crutches in that lora but the latest version just a pure luck with deeper colors and without artifacts on cfg++
>>29043 Same, I want to eat pizza and play vidya with her. >>29045 >but the latest version just a pure luck with deeper colors and without artifacts on cfg++ I've been using 0.75-based one (v3.2) tbh, I had some issues with 1.0 lora version when I was genning on v29 and 0.75 felt like it had the best compatibility with any vpred models I've used. Someone mentioned that v29 was trained from 0.75, but I don't know if that's true or not.
>>29028 i used these wildcards. they should be in OP for convenience: https://rentry.org/NAIwildcards i'll share a few of the longer wildcards i had to curate though. https://danbooru.donmai.us/wiki_pages/tag_groups
>>29046 3.2 was just half of the time baked than 4.02, only noticed afterwards that some styles was not showing up, strange that it works the best, when there is literally dedicated version for v29. I liked the most next 4.02 version and made the latest dataset on top of that. IIRC when I was testing noobs, only 1.0 showed up as correctly converged vpred, or maybe also v29. Not just from meme bg test, but more importantly when prompting white skinned, white haired, white dressed char to be far away from the camera on plain white bg and not have space to be puffed with random shit or shifted bg color. Similarly to choco with black hair, black dress on black bg
>>29048 >when there is literally dedicated version for v29 I currently use 1.0+v29 I'll try the loras again, I haven't been genning too much lately and didn't test this stuff too hard
(5.68 MB 4032x4032 file.webp)

(7.41 MB 4032x4032 file.webp)

>>29041 >gen multiple poses/outfits for a given character KohakuBlueLeaf trained a prompt to prompt LLM for that https://github.com/KohakuBlueleaf/z-tipo-extension It can generate an entire set of outfit from just "alternate costume" or an entire sex scene from "sex" Sadly the webui extension for it is outdated
(282.29 KB 1344x1728 00003-3102262158.webp)

>>29052 Thanks, I'll give it a look.
(397.93 KB 1686x923 Capture.PNG)

>>29052 >>29053 (Me) >Sadly the webui extension for it is outdated The github link you gave me works and seems to have been updated recently, what's outdated, exactly? Have the devs just not ported over new features to webui?
>>29054 >>29054 Oh sorry I ran into some weird problems led me to believe it's imcompatible with reforge, probably specific to the reforge commit I was using or my env, the extension was updated 2 days ago
(355.62 KB 1344x1728 00004-3005049992.jpg)

>>29056 Oh, I see. Don't worry, things also break randomly for me in reforge. Regarding the extension, the prompts it comes up with aren't very good, to be entirely honest.
(289.05 KB 1344x1728 00027-1583679284.jpg)

(301.67 KB 1344x1728 00035-1939488725.jpg)

(275.24 KB 1344x1728 00046-520760656.jpg)

(201.99 KB 1344x1728 00048-2059586672.jpg)

For your own safety, you can't see the entirety of Ononoki's naked body at once. You'd die from the eroticness. Instant death.
why is webp still not allowed on 4chan
>>29061 Cursed.
>>29049 I really don't like this mix, it just overexposing whites, not as bad as some other mixes, but still https://files.catbox.moe/9gvee3.png Don't know also what to say about your style in particular, it looks pretty much the same across models and loras https://files.catbox.moe/k5w7kr.png pretty strong style Any other thoughts from tests?
>>29065 I tried a bunch of v29 merges and they all felt artifacty/smudgy to me and white was off no matter what settings I used. The only 1.0 vpred merge I've liked was the one linked in >>28891
>>29042 which lora?
>>29066 Yeah, that one is better at handling whites https://files.catbox.moe/fm7o57.png pretty kino result
>>29067 https://civitai.com/models/723360/ai-styles-dump-illustriousponynoob i've mostly been using this https://civitai.com/models/723360?modelVersionId=1155642 but you probably should go for later versions. >>29065 >>29066 1.0+v29 indeed looks pretty bad in that example but I've been reasonably happy with it for my gens. I'll retry the lora and try the vpred_fix model though, later
>>29070 AH WELL EVERYBODY'S HEARD ABOUT THE BIRD
Can anyone explain to me the intended way to run reforge on Linux before I bang my head against the wall any further? I'd usually assume you run one of the shell commands, but considering https://github.com/Panchovix/stable-diffusion-webui-reForge/commit/314b3ae5de20338b1b7ea7b120191452ee2faabf, I'm genuinely confused. Also the webui-user.sh script does literally nothing.
>>29070 huh, I remember seeing her a long time ago. Does she really have that many entries to be recognized?
>>29072 >webui-user.sh Unlike with windows, it exists only to set up environment variables during launch. I'm running it with just ./webui.sh, don't know what is about torch 2.7+cu128 there, but you can set up any version of torch inside venv if you need
>>29074 Thanks. I'll try to just pip install some sort of eenie meenie torch version and then see if reforge just works.
What was the name of the extension used if I wanted to generate two prompts in one image or two completely different images in one?
>>29080 a daring synthesis
(2.04 MB 1600x1200 catbox_moxkvk.png)

(2.27 MB 1344x1728 00029-641581694.png)

>>29075 (Me) Sorted it out. The only relevant bit of advice I have for anyone trying to run reForge on some other goofy distro is "some of the errors are wrapped, so create a test.py and try to directly import torch and check for CUDA to get actual errors."
>>29080 the ghibli style doesn't really do it for me but there's something nefariously erotic about seeing henrietta in it
>>29083 cute pantsu. lower the CFG or strength of whatever lora you are using btw. her cheeks are kinda scuffed.
(1.68 MB 1344x1728 00034-971718895.png)

>>29085 >lower the CFG or strength of whatever lora you are using btw. her cheeks are kinda scuffed. Ah, thanks. I think I just rolled a weird gen, though, cfg 3 seems fine.
>>29087 >it has metadata silly me. very cute.
>>29088 Me in the bucket (I'm Solid Snake).
>they refuse to announce a release date on the discord >they do it through a payment processor update popup What the hell is wrong with analtan? also paddle dropped all generative ai services because muh problematic content that they've known about for ages
>>29090 >you can't generate certain kind of content with bing or imagen, it's unsafe ok >even if this other company allows you to it's still unsafe >we won't even let you pay so that you can't actually generate anything i fucking hate this and everyone involved in it, same with that entire dlsite bullshit being told what i can or can't spend my money on when it comes to fiction and entertainment makes my fucking blood boil with rage
>>29091 just remember that it's not about not letting you pay for porn, it's about not letting you pay for porn (((they))) didn't make/don't approve of/can't use as propaganda this is the full statement from paddle btw https://x.com/andjdavies/status/1836292129666056631 at least analtan already found another processor
(1.96 MB 1600x1200 catbox_li5sz3.png)

(298.49 KB 1152x896 catbox_ulvj3i.jpg)

I was trying to reproduce an image I picked up on the internet, but I was wondering how these particular toys are output...? It was my hobby to put two together.
Why does hi-res fix add a blur filter?
>>29094 It doesn't, what settings are you using?
>>29065 Box for 3rd image onegai
>>29094 Are you using latent upscaling? Use something like AnimeSharpv3 with ~0.3 denoise.
>>29094 >>29097 Note: latent for vpred models stops adding the blur around ~0.8 denoising.
(1.22 MB 896x1152 catbox_wp9lx5.png)

(2.51 MB 1344x1728 catbox_asxwuk.png)

>>29098 Latent also destroys image composition beyond 0.8 denoise.
>>29099 nta but you can use tile controlnet tile for that that was my preferred upscale method (latent 0.8 denoise + tile controllnet) during 1.5 times but I never bothered while on noob. might be worth considering
explains a few things
>>29101 based backbone haver
>>29101 How is v4 full different to v3 when it comes to nsfw? I assume they're still pruning realistic shit right
>>29103 I was gonna make a joke along the lines of "(Do not reply to this post. Contain the urge to reply because it will be deleted later and you will be deleted along with it if you reply) levels of cunny" but then i remembered that that's actual cheese pizza and it stopped being funny
>>29104 damn, when was THIS implemented?
>>29105 Pretty sure this board always had shit like this, LMAO
>>29106 it did but i'm sure that it got corrected to "no-no model" back when it was relevant instead of whatever this is
>>29103 half-informed lawyers panicking that the model wasn't made by an outside group that blame can be scraped off onto, would be my guess generative ai in general is still in the "running around in an open field below gathering stormclouds, praying for someone else to be the one who gets struck" phase of developing case law
>>29107 nvm im slow
(1.11 MB 896x1152 00077-1809270274.png)

Wait does noob produce real japanese text (pre-canned sentences probably) or is this suspiciously good looking gibberish?
(3.51 MB 1344x1728 catbox_a76dux.png)

(3.69 MB 1344x1728 catbox_h3yufu.png)

(3.31 MB 1344x1728 catbox_33vvqk.png)

(3.13 MB 1344x1728 catbox_feneuy.png)

>>29110 that says "cho!" which is more of a sound effect than a sentence
>>29105 >>29106 >>29107 >>29109 That is OLD -OLD- A relic of an old time where there was a CSAM model being posted on /hdg/ and was there so people didn't post it here or bring that shit here. It was never removed because everyone involved just forgot about it so you're probably the first guy to trip into it in like a year or maybe over that.
>>29112 like i've said, i remember it getting aliased to no-no model and not this thing
Welp i didnt expect much from v4 but its certainly underwhelming
>>29115 >>29114 I was going to post about this but got sidetracked. Yeah, pretty sure they are aware v4 is bad and were shilling v5 while saying v5 is a bigger leap than v2 to v3 so who knows if they'll live up to the hype they're setting up.
>>29114 Does the bit about associating tags with specific characters actually work reliably?
>>29116 supposedly v5 is still coming before the end of the year so who knows i'll give v4 a try once i get some rest i'm more concerned about the dataset thing and disabling tags tbh, wonder how the data still slipped in despite the combing and retraining also no SMEA apparently because "it never worked properly"
>>291 >>29117 Does work decently with two subjects but as long as there are more then it does whatever it wants, also it feels super stiff compared to what little tags gave you on v3 which might have to so with the dumb decision of adding natural language
Also the worst part at least in my opinion is how bad the artist mixing works and if you try to enforce it by adding brackets it adds a lot of artifacts...
>>28403 >stealth png was off shit. it's gone. 'pretty sure i just copied shutenanons nai prompt onto better days though.
Verdict on v4? halfchan is obviously no help
>>29123 Underbaked, nerfed, and they deleted the furry artist tags again.
>>29124 nerfed in what way? I don't care about furry tags
>>29123 I don't like images I've seen so far. It looks noisy, mixing works poorly. Text is fun but doesn't feel like THE killer feature worth the hassle. If I was NAI-only I'd prefer them to update v3 with latest characters, to be honest. Maybe I'm mistaken and someone will find a way to prompt amazing shit from it.
>>29123 Style mixing is dead in the water and kills a lot of the fun for me. I otherwise like it and think its a fine model.
(1.00 MB 832x1216 catbox_nzb1ky.png)

>>29127 I think there are ways to mix styles. but imo individual styles work much better and hands are finally improved. i tested it out with the remaining points i had saved up. it might be worth resubbing
>>29128 NGL that looks pretty bad.
(1.74 MB 832x1216 catbox_3zksp4.png)

>>29129 similar prompt on local. naisisters what happened?
>>29130 I like this one better but I'll have to give NAI the win if the teal outline is on every image in the grid.
(1.14 MB 832x1216 catbox_818vmy.png)

>>29131 V4 was VERY overhyped it seems. mogudan looks horrible, for example.
>>29132 Does v4 have a newest tag like noob you can mix in to filter out the really old mogudan stuff? Mogudan's old art has awful res on danbooru: https://danbooru.donmai.us/posts/34914?q=mogudan+
(1.14 MB 832x1216 catbox_818vmy.png)

>>29133 switched to UC heavy preset and mogudan looks significantly less derpy. i'm almost out of Anlas: and i'm not too sold on V4 yet tbh. obviously you can't get a nuanced opinion from halfchan
(1.22 MB 832x1216 catbox_a77fa0.png)

>>29134 pffft wrong attachment
>>29134 >>29135 I see. Funny that it's doing the white border on both images. >obviously you can't get a nuanced opinion from halfchan Youtuber niche waiting to be filled.
(1.20 MB 1216x832 catbox_utigtj.png)

>>29136 again I'm really torn on V4 because it can go from producing decent results to making 1.5 tier slop
(1.10 MB 1216x832 catbox_1gt6x9.png)

(1.11 MB 1216x832 catbox_6pg2ab.png)

there's just something "off" about V4 that not even the character and text prompting can overcome
>>29139 Does feel that way. I have ~9k anlas left over from when I signed Opus a few months ago so just post it on the thread if you run out and want someone to test some random styles/poses.
>>29140 most notably, it's hands. they're very artist dependant
>>29141 I can kinda see it, yeah.
>>29142 Oh yeah, it just can't do the double v with this style. It might be just struggling with more delicate hands and succeeding with simpler, "meatier" hands.
>>29143 >It might be just struggling with more delicate hands and succeeding with simpler, "meatier" hands. Nvm, it seems to be just random from artist to artist.
if ya gonna test v4 try using your good v3 prompts and then put images side by side, that would be interesting
I'm noticing that it doesn't really fuck up the number of fingers, it just draws really scuffed hands. >>29145 Sorry, don't have those, didn't use v3 much. Just post the pngs and I'll burn the anlas though.
v4 is kinda soulful when you're trying to gen stuff that isn't porn.
>>29139 >there's just something "off" about V4 that not even the character and text prompting can overcome For me it's the noise. It seems to be present in like 80% of gens.
yay it can do gestart not as good as local with lora
>karras >exponential >polyexponential NAI stop doing goofy shit and implement some better noise schedules.
mess around with cfg rescale, it helped a bit with artists that looked scuffed at first
NAI v4 is pretty good at portraits, actually.
>>29153 Seems like its pretty good for flat color stuff
Which cunny artists look good in v4? I feel like they're all either underbaked or fried and there's no inbetween
>>29154 ...sort of. >>29155 I'd still just use noob for that tbqh.
>>29155 fizrot is actually accurate but the eyes always look like they're glowing and you can't solve it by negging glowing eyes because nobody tagged them properly lmao
>>29157 bright pupils?
>>29158 super bright and overly saturated irides compared to the rest of the image (as if they were glowing) also no actual pupils for some reason
>>29159 i meant negging "bright pupils" cause it's a tag, i've had the issue you described with some artists on local and putting that in negatives helped
>>29160 ah, lol, one sec
>>29161 >>29160 nope, still too bright and still no pupils
NAI really went and made a model that's best at genning profile pictures and I can't stop laughing at this.
(1.39 MB 832x1216 00035-1798953701.png)

>>29162 i see, but i suppose it is kind of a "feature" of the artist, isn't it? seems pretty similar on my local+lora pic and here too: >>28361
>>29164 It's definitely a feature, maybe it looks worse because actual fizrot eyes usually have a more pronounced brush stroke effect when he doesn't draw the pupils but on V4 it looks like artifacting instead. I'll write a big wall of V4 complaints nobody's gonna read later.
>>29165 either way shit isn't tagged properly so it's once again a tagging issue at the source
>>29165 I'll read them but primarily so I have more ammunitio ln to anti-shill NAI.
>>29167 >ammunitio ln to anti-shill NAI. meds though
>>29168 sorry my hands are shaking erratically due to the meds
(1.31 MB 1216x832 catbox_u1pbhj.png)

yeah V4 can do funny stuff. but it's fundamentally a downgrade from even local. i'm glad i had the anlas so i saved the 25 bucks. all signs look like it was a half baked model that was overhyped because they were bleeding subs with how long it was taking
>>29126 it took a few days before i started to get anything notably good with v3
>they just extended the cfg scale slider to FUCKING TWENTY (20) BUT WHY he's pogging tho look at him go
was gonna mention this in the wall of text but this is genuinely fucking dogshit and completely unacceptable they used to tell you to just neg white background and something else and called it a day but it seems like it's such a problem that they had to intervene directly also no refunds if it happens after using anals for whatever reason even if all that compute gets used to make a completely white image as if 1.4 just NaN'd at the very last step of hiresfix
>>29173 and supposedly v5 will still have this for whatever reason even if it's "to a much lesser extent" v3 can shit itself too but it usually still spits out a fucking image one way or another, v4 just says fuck you lol here's nothing
they really should just scrap it and just update the v3 dataset instead
>>29174 they should've just baked v3.5 on the same exact settings as v3 while working on what will hopefully be v5
>>29165 I love reading people's seething. >>29173 People who upload blank pages to danbooru or approve them and then forget to add them to pools go to hell when they die. https://danbooru.donmai.us/posts?tags=blank_page&z=1
>>29177 Who am I supposed to seethe at here, DB for being dogshit or Analtan FOR NOT REMOVING SOMETHING THIS FUCKING OBVIOUS BEFORE HITTING THE RETRAIN BUTTON
>>29178 Probably anlatan because I'd imagine they removed these on v3.
>>29179 v3 and now v4 were trained on posts tagged with ai-generated so I dunno.
honestly i have a bigger problem with their dogshit attitude than i do with the failed bake that is v4
>>29181 if anything, it shows that v5 isn't going to be any better. They have their heads so far up their own asses they cant see straight
>>29182 >They have their heads so far up their own asses they cant see straight i half agree there, i think deep down they know v4 is mostly a failed bake my only hope for v5 is that it will actually be what v4 was supposed to be. as 90% custom architecture it's not BAD, it could've been so much worse but i'm convinced they didn't actually finish working on it and that they definitely fucked something up during training i think they wasted too much time and compute on coping with aetherroom (lol at them finetuning a llama model that immediately became massively outdated) and only started working on v4 (read: rushing it) when they realized they hadn't released something new in over a year on the other hand i h(c)ope that their heads actually are that far up their asses and that they'll think that v5 will be so good that releasing the weights for v3 won't come back to bite them but i already know they won't do that (let's be honest it's not like local bakers could actually improve it either)
also jesus christ they need to tell occultsage to shut the fuck up or straight up fire him because the guy isn't just completely blind (in the same way halfchan anons were unironically comparing pony to v3 with the same prompt and styles and going "we can do that too, they're the same picture!") but he's completely fucking retarded and incompetent too
>>29184 forgot to attach the image
>>29185 are they fucking for real
>>29185 He's taking the piss, right? There's no way he doesn't remember what a drawing with colored pencils looks like from school.
>>29186 it's occultsage, same guy who told people to describe the style in prose and then to ask gpt to make you a prompt based on the style i'm convinced they haven't fired him just because he's the only textgen dev left on the team
I feel like nsfw is way underbaked in v4 things like nipples,dicks, cum, etc have like 0 detail in the case of nipples, sometimes they're just non-existent on certain artists
What was the name of the extension that generated the background and the character separately? I hate how schizophrenic the backgrounds are in my gens and really want to fix that
>>29189 literally feels like like they hastily added NSFW data to the preview
Someone remember what was the cunny mix this guy using? https://www.pixiv.net/en/artworks/120814370 I remember seeing similar pictures with metadata here or on halfchan, but forgot where
I tested out V4 and while its severely disappointing, I remembered that one thing it exceeds at when it comes to some artist styles is an accurate texture. So going back to noob made me feel like the model lacks the ability to capture a lot of textures that may contribute to a better accuracy for a lot of artists, at least without loras. Would training a lora help with that? I wouldn't really know since I haven't been able to train one, I lack a strong gpu.
>>29192 something something key_(kagibangou), naruko_hanaharu, kishida mel, cutesexyrobutts, meis_(terameisu), lasto, murata range, gomennasai, aak, muk (monsieur) These were in all posted images and genned in NAI but I've never figured out proper weights to translate to local. Technically there's also trained lora here https://civitai.com/models/723360/ai-styles-dump-illustriousrouweinoob tagged as "amamizore" but it barely works for me. And there's an entire model trained with this style https://civitai.com/models/916130?modelVersionId=1025392 Didn't really like it but it kinda works.
>>29194 Thanks, I think I found it >Technically there's also trained lora here I know, this is why I actually want to remove this gan upscaled shiny slop from dataset and replace it by a native style from v3 Is this even real to make it with v4? I've tried and it outputs something like pic 2
>>29192 >>29194 That was probably me posting it. Although the one you linked looks a bit different. key_(kagibangou), naruko_hanaharu, kishida mel, meis_(terameisu), lasto, [[murata range,gomennasai, aak]], [[[[muk (monsieur)]]]],
>>29196 That looks better, even second image from v4
(586.47 KB 652x1082 1740894014405415.png)

Just in case.
(1.82 MB 1920x1080 NAI V4 Full.mp4)

Here's my long ass "V4 *so far*" writeup that nobody asked for, I took my sweet time because I watched Dosh talk about trash for 50 mins first in order to better prepare for this and then I got distracted and watched Tenma get fucked by Sephiroth again for hours on end If it's too long for you and/or you don't like my blogposts and/or tangents don't bitch and just hide the post >The Good (surprisingly many but not enough to outweigh the negatives) 1) Styles are more accurate (and so are characters but and still need less images than V3 to work (translation: I only cared to test Shoebill EX (yes, there's a Shoebill alter that's pretty neat but is too different and nowhere near as good as the original and it also doesn't really make that much sense?) but she works well enough with just the character and copyright tag and only has around ~30 images, YMMV) 2) The inpainting model works well, they didn't fuck that up. It works REALLY well with the new inpaint focus area thing but that's also available on V3 apparently. Mixed feelings on the not-really-circular brush (also on V3) 3) NSFW concepts work way better than V3 on average 4) Single styles work really well, so far I've only found one that doesn't work anywhere near as well as I hoped but whatever. Your mileage WILL vary 5) 512 tokens are a godsend, it's nice to just export booru tags 6) The regional thing works well, it's annoying to use but at least with two characters (± 1boy pov) it just works™ and it's worth using just to specify the characters' positions, will test 3+ later. The source#target thing is annoying as shit though and I feel like this is VERY hit or miss for anything more complex than looking at/hugging >The "who the fuck cares?" 1) NLP. Like I've said, it's nice when you need to fill in some gaps that tags can't fill but 90% of the time it's a case of "I wish shit was tagged better at a booru level" and not "I wish this hyper-specific tag existed at all" 2) Text. Again, faster to do in PS and unless I have a massive skill issue there's no way to actually have it blend in on surfaces instead of having it look like it was slapped on in MS Paint. At least it works better on Flux 3) The 16ch VAE is mostly a meme. Eyes and details ARE better on average compared to V3 but we're talking about 4070 SUPER vs 4070 Ti SUPER kind of gains in the best of cases. Very small details like the text on Shishiro Botan's HOLOLIVE strap are slightly less fucked but I expected the 16ch VAE to more or less fix stuff like that >The Bad 1*) Emphasis does fucking nothing the vast majority of the time, I don't know if this is because DiT is inherently dogshit or if it's because they really did train with emphasized tags on somehow. >The Ugly 1*) Anatomy (and backgrounds) seems to be COMPLETELY artist-dependent now (instead of mostly) and for some reason hands are just scuffed in general, you don't get extra fingers that often but the hands are just... "drawn" badly. 2) Related to the issues* above but emphasizing size tags (eg huge breasts) does ABSOLUTELY NOTHING IF YOU'RE USING AN ARTIST STYLE. You can test this by simply reusing the same seed. Without using artist tags it actually works to a degree but this brings us to the next issue 3) HOLY FUCKING MOTHER OF ABYSSORANGEMIX1, WHAT THE FUCK IS UP WITH THE DEFAULT STYLE? It's not ALWAYS AOM1 slop but it tends to gravitate towards that, even more so if you use one of the completely broken* >The Dogshit 1)* medium tags. Traditional media, marker, millipen, etc etc. These are nearly completely broken. Some will actually change the style but not to what the tag says. Oil painting seems to work to a degree, woohoo 2) Mixing, obviously. Finicky, unstable and inconsistent IF you can even get shit to mix to begin with. DiT was a mistake 3) Weird jpeg-like-but-not-quite? artifacts everywhere, sometimes it almost looks like it's trying to imitate the grain effect used by some artists regardless of the artist tag. Some anons also reported incredibly bad frying on top of the artifacting but I haven't seen it yet 4) Shots wider than full body/tachi-e (sorry, "character image") images are an absolute crapshoot, maybe even worse than V3's but I'm not gonna bother checking 5) Some combinations of character tags and framing tags (and/or resolution too, especially square) will occasionally/consistently (it really depends) output REALLY blurry images. This was worse on the preview but it's still frequent enough on Full to genuinely be an issue. It's not even an unprompted depth of field effect (which happens more often than you'd think) either. All framing tags seem to do this to a degree but it's extremely common with "close-up, portrait," even with artist tags 6) The fucking white gens. According to finetune removing the "blank page" images from the dataset wouldn't solve the issue because "it's probably not the culprit and removing them would remove this (blank page in negs) workaround" even though it seems like that's exactly what it is. Either way this is completely fucking unacceptable, imagine if V1 (on their site) actually NaN'd out like the shitmixes did on local. The fact that they went from telling people how to fix it to actually having to intervene makes me think that they don't believe their own "oh it just happens sometimes, not a big deal" bullshit >The Lovebite Dancin' in the shadow with a dark smile I probably forgot about some stuff but overall: see the attached video Birdschizo out, I'm going back to playing TXR 2025.
>>29199 >DiT was a mistake isn't this still a mostly unet model and the mixing shits itself due to t5 te?
>>29200 I don't remember the exact wording or when they said it but either kurumuz or finetune (probably kurumuz) said the model is "90% transformers at this point"
agree with the big wall of autism above hits some concepts v3 and local struggled with, it can do portal limb abduction and almost extend that to gap limb abduction hope they unfuck it
OH YEAH BY THE WAY THEY FORGOT TO CAP THE CFG WHEN THEY EXTENDED IT TO 20 SO HERE'S 10000 GUIDANCE LOL
>/hdg/ is all swarthoids arguing about interracial porn wonderful
Is there some general rule of the thumb to get only the desirable number of characters? Not sure what went wrong with my tags but I'm getting random characters in the background. Usual combination of 'duo, couple focus, 1girl, 1boy' was working well but for some reason I do get sometimes super deformed chibis peeking in the background. I'm not really sure what went wrong.
>>29205 Just pollution from artist tags. Happens both on noob an NAI. "chibi inset, crowd, people" in negs always cleans this shit up for me.
>>29206 Gotcha, thank you! I think this reigned it in a bit.
>>29200 >>29201 someone found the screenshot(s), looks like it's not DiT but basically behaves just like it? that sounds fucking retarded and i included an extra (the first one) what the fuck does he mean by that?
>>29209 unless they release weights we will never know what they're actually doing.
>>29209 They're doing whatever scuffed shit seems to work, I guess.
>>29211 yeah, the key word there is SEEMS
man nsfw abmayo is really scuffed on v4
>>29199 their denial and awful reaction to criticism along with ugly aomslop gens from occultretard is probably biggest issue with all of this for me. failbakes can happen, especially when you are going for a big gamble such as your own foundational model but man they aren't handling this well.
(2.33 MB 1600x1200 catbox_7rxegx.png)

What NEGATIVE is needed to prevent two anuses?
>>29214 Noob has basically fulfilled the promise of what V4 should've been. (SDXL finetune with updated danbooru + furry dataset). it's insane that they didn't just take the easy route
Something with their resolution is fucked too I noticed portrait mode doesn't have nearly as much issues as landscape does, in landscape there's like a vaseline filter over every gen or its just fried and has artifacting but, that seemingly doesnt happen at all in portrait
>>29216 at least they should have released updated sdxl model while they're still fiddling with their experimental bake. that would have been enough to retain their customers while they're trying to do the big thing.
>>29214 >their denial and awful reaction to criticism [...] is probably biggest issue with all of this for me Yeah, ditto. A failbake can be forgiven but man their PR situation is completely fucking hopeless. V5 needs to be SOTA if they want to get even the slightest chance to rebuild all the bridges they're burning, nobody's buying their shit.
>>29217 I noticed that this >>29199 >5) Some combinations of character tags and framing tags (and/or resolution too, especially square) will occasionally/consistently (it really depends) output REALLY blurry images. This was worse on the preview but it's still frequent enough on Full to genuinely be an issue. It's not even an unprompted depth of field effect (which happens more often than you'd think) either. All framing tags seem to do this to a degree but it's extremely common with "close-up, portrait," even with artist tags is even more common at 1024x1024 and in landscape. Portrait in general seems to be the least fucked resolution but they're ALL fucked (meanwhile on V3 I could swear that the square res was the best of both worlds unless you needed very specific framing)
Honestly, some styles being (more) accurate is a double-edged sword in some cases (ie fizrot)
Also I feel like if you're doing anything more complex than 1girl, standing every sampler that's not Euler A just shits the bed catastrophically
>>29199 I've only tested for a brief period while the free instance from 4chan was working but I agree with most of this Tried a few scenes which are broken / unstable on noob, didn't see any improvement on NAIv4 >>29202 >a chain running from her clitoris to her collar Impressive if at least semi stable
they need to fire this faggot immediately
>>29227 is he mentally retarded or are they just using him as a fall guy lmao
>>29228 the r*ddit mod attitude, his takes and the fact that he has his own emojis including fucking soyjak :OOOO faces leads me to believe that he's genuinely mentally retarded
>>29228 >>29229 even finetune is tired of his shit
>>29227 >>29227 He's actually making v4 sound even worse than it really is lmao.
>>29227 okay, that makes so much fucking sense now jesus christ
>>29232 >2.0 code for stability I wonder if he's involved with their decision not to use vpred+ztsnr and cursing the open weights community for 2+ years
>>29233 there was no ztsnr paper by the 2.0 time, it came out almost before xl drop, 2.0 model, however, used vprediction by itself, but without a proper fix. They probably could retrain xl properly at the moment of ztsnr paper was out, but haven't bothered with that and dropped the weights, as long as I remember that
>>29232 He's making it sound like some research contribution but all he did was write some dev tools. This is their GitHub: https://github.com/wbrown https://github.com/Stability-AI/api-interfaces https://github.com/Stability-AI/stability-sdk-go (I was going to post this on 4chins but some filtering is happening with my IP range currently, feel free to post there)
>>29235 so basically he's the equivalent of the blizzard nepo baby without the nepotism
(1.11 MB 1216x832 rescale 0.4.png)

>>29235 https://github.com/Stability-AI/stability-sdk-go/commit/19432f75d63e42df3cc78db2aac7608a59d83533 >fmt.Println(fmt.Sprintf("WARNING: %v", decodeErr)) I wanted to look through his code for something goofy but I didn't think it'd be this fast. It's still like that https://github.com/Stability-AI/stability-sdk-go/blob/main/cmd/interrogate/interrogate.go#L24
>>29240 Some of the poses and multichar are impressive but wow, v4 feels so aesthetically lacking compared to v3. Some of your Shuten gens got me close to subbing once or twice because they looked so clean, everything I've seen from v4 is missing the sauce.
(1.22 MB 1024x1024 catbox_ws4ez2.png)

>>29241 yeah, V4 just lacks that "aura" that made V3 a big real
(814.76 KB 1920x2176 comparison.jpg)

>>29241 Pretty much, this one isn't optimized for v4 but it still encompasses a lot of problems I have with it. It seems to lack a lot of color shading and clean transitions between colors. Everything is very sharp(?) and colors seem very flat. Imo I think v4 is very good for artstyles that are like, intended to be very high saturated colors with thick outlines if that makes sense. Artists that don't fall under that come out either incredibly underbaked or outright fried and there's very little in between. I think most of the "clean" feeling came from SMEA effectively blending the artstyles together in v3.
>>29243 >I think most of the "clean" feeling came from SMEA effectively blending the artstyles together in v3. tbh I don't agree here, imo SMEA wasn't always beneficial on v3. it's great when it works with the artist mix but v3 could look nice without it too. v4 is very lacking so far, yeah.
>>29244 Fair, could be more of an artist dependent relation. v4 feels more like an alpha version of a model rather than a full upgrade. Posing, composition, and even inpainting are genuinely good but, without the aesthetics there's not a whole lot of reason to use it over local for that.
(2.39 MB 1600x1200 catbox_haapw4.png)

(1.83 MB 1824x1248 catbox_7pqyd4.png)

Halfchan has devolved into whining about black dicks to the point i still have no idea if V4 can be tardwrangled or if it's not even worth bothering
>>29247 people seem to get cool gens from it composition-wise but the biggest detractor that v4 just looks crappy.
>>29248 I've seen decent gens from V4. but I genuinely don't think 25 is worth it just to learn it when Noob has surpassed V3
>>29249 i would strongly prefer v3 to noob but considering v3 costs money and noob has all the local tools i guess you're right
>>29238 >Align Your Steps 11 where do these new schedule types come from?
>>29251 research papers
>>29247 >Halfchan has devolved into whining about black dicks what else is new anyway nyesn't I'll just copy-paste my halfchan post "let me put it this way, it's 10x better at being a sfw/shitposting model than it is at being a nsfw one despite the fact that it can do nsfw concepts and poses that you wouldn't even dream of doing on v3+inpainting" >>29245 >v4 feels more like an alpha version of a model rather than a full upgrade. it genuinely does, yeah the model isn't underbaked (the curated one is though), the architecture is forgot to put that in the big wall of text
(1.66 MB 1824x1248 catbox_issp7z.png)

>>29255 Guess I'll just wait before I pull the trigger on experimenting with V4. (American food prices are rape)
>>29256 write down your trooncord burner and i'll let you use my api token
>>29257 just "sdmsakuya"
(895.49 KB 832x1216 catbox_2h7828.png)

(761.93 KB 832x1216 catbox_ro3dc6.png)

V4's text is really good at shitposting
(1.24 MB 1216x832 catbox_5brzfk.png)

anatomy is a bit scuffed. but putting "ai-generated" in negatives helps a lot
>>29260 anatomy is entirely dependent on artists now, that's why at least with v3 things couldn't get too too weird
>>29261 >at least with v3 things couldn't get too too weird (body horror and ghost limbs were still a thing but i feel like it was more consistent overall)
(1.15 MB 1216x832 catbox_v8vwjg.png)

>>29261 so far i've had decent results with optionaltypo
>>29263 v4 is mostly ok with flat styles or styles that aren't too detailed or don't have complex shading or any kind of actual shading really it kind of reminds me of eps vs vpred where eps models' shading always looks like a binary sort of deal and there's no real transition regardless of the style on v4 it's just artifacts and other weirdness
(527.52 KB 4000x2737 catbox_ihoekp.jpg)

i'm starting to warm up to V4 in some aspects. you have FAR more control over the guy than you ever could have on local
man, I guess local is allowed to be just okay for a few months before anlatan releases a half baked model that still shits all over it. maybe i should just kill myself at this point.
>>29267 v4, even if salvageable, is still too scuffed to say this
>>29265 >>29266 it's not too hard to get it to spit out something decent to good and to a degree you can enhance 1x/1.5x the artifacts away but it IS too hard to get over the "this looks decent but it would look great with my old style mix" mentality if v3 didn't exist v4 would be genuinely impressive despite all the shortcomings
>>29268 people are already figuring out how to work around the scuff and their inevitable 4.1/4.5/5.0 will be a solved problem. i could spend hours writing an essay about why this is bad, why you shouldn't give them money, how i could buy 50 years worth of subs right now but don't, yet it doesn't matter and everyone would think I'm a schizo or trolling anyways. it's just so fucking grim and the whole situation makes me sad.
>>29270 >people are already figuring out how to work around the scuff by all means point me to their findings because i can't fucking figure it out and no amount of unscuffing will solve the main issue (artist mixing working 1/1000th as well as it used to)
>>29270 i think you're taking this shit too seriously it's not some food subscription NWO nightmare, it's an ai service you pay for when you want to have fun and forget about it when you're no longer interested if you don't want to spend your money due to some conviction it's fine too, it's only unfortunate if you want it but can't afford it
>>29272 >it's only unfortunate if you want it but can't afford it nta but it's a real shame they don't have a cheaper imagegen-only plan, most of the compute is used up to host and train the llms which most people don't even use or it's 100% wasted on trying to salvage aetherroom (i say this while genuinely contemplating paying for shit on openrouter because free proxies are dead? and i can't find a paid one i could even think about spending money on)
>>29274 your wife is so talented...
>>29275 She really is the best, cute wife made for sex
>>29274 I'd honestly rather go back to the way less accurate but way cleaner styles of v3 than v4's more accurate but artifact-y ones
>>29277 than be "forced" to put up with* accidental ctrl-x
since halfchan is dead i'll just ask here do any of the textfags here have a spare token for some still (/or know of any) living proxy?
>>29279 remaining proxies are pretty closed and quiet as far as i know. if you have a burner google account you can use free gemini through aistudio dot google dot com.
>>29280 i don't, RIP guess i really should've bought my way into a proxy after all
HOLY SHIT THEY INTRODUCED AN ACTUAL IMPROVEMENT that also works on v3 genuinely fucking huge, both options (don't overlay and get noise or overlay and be forced to enhance at 1x/1.5x to cover it up and possibly lose details) were fucking garbage
who killed 4cuck
Kinda funny how Illustrious 1.0 feels better than 1.1 because 1.0 can be downloaded and finetuned vs 1.1 that can't The illustrious guys think they're NAI but the strength of local has always been the finetunes and the loras
(165.21 KB 425x369 yougotme.png)

christ i made the mistake of checking 4ch/hdg/ for a while and, console wars aside, I cannot believe that highlights faggot is still going. how does one man have enough schizophrenia and free time to always be there to make the thread worse?
>>29287 I started lurking again few days before v4 release and it feels like the whole thread just runs on pure schizophrenia now
>>29287 shitpost hard enough and he'll only do one highlight.
>>29287 sounds like someone didn't make it into the highlights
>>29285 >Kinda funny how Illustrious 1.0 feels better than 1.1 because 1.0 can be downloaded and finetuned vs 1.1 that can't sir your "i fought in the pony-nai wars and lost" medal is rusting
the character positioner is pretty nice actually, a lot better than the hacky shit i was doing with high denoise img2img of mspaint sketches
>>29292 I tried to use it to gen futa on male and it didn't fucking work. Ideas?
(3.00 MB 1624x1840 catbox_ga7grx.png)

(3.12 MB 1624x1840 catbox_7yk5xz.png)

(3.06 MB 1624x1840 catbox_kzo786.png)

(2.87 MB 1624x1840 catbox_2z4e1j.png)

>>29293 you gotta support it with all the proper tags and don't duplicate between fields


Forms
Delete
Report
Quick Reply