>>28235 Thanks for baking

https://github.com/neggles/sd-webui-stealth-pnginfo please use onegai

>>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

>>28042 >asuka

New bread, more beach wives

>>28239 nta but this extension allow to store metadata https://github.com/catboxanon/comfyui_stealth_pnginfo

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

>>28239 i do some of that too but i always plug it into a zero denoise img2img at the end to paste the metadata back on because i've lost stuff i liked before that way

>>28237 catboxes?

someone stop this sick criminal

>>28252 https://files.catbox.moe/b9mmey.png https://files.catbox.moe/43vozt.png https://files.catbox.moe/4hazo0.png https://files.catbox.moe/3j89ok.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

>>28269 cute

>>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!

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

time to cook a new prompt

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?

>>28292 https://files.catbox.moe/ojz681.png https://files.catbox.moe/ea78nj.png https://files.catbox.moe/9him2u.png

>>28202 >>28221 >>28223 I'm quite late but DAMN any catboxes? Please.

>>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.

>>28296 high quality

>>28296 yeah this is peak SEXO >>28297 very cute

>>28283 by order of preference 1.cute and funny 2.wholesome and silly 3.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

fucking loves wheat

>>28303 the image quality here is a little..... grainy jokes aside it's fitting, very cute

>>28302 Seems alright to me

>>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.

>>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.

>>28238 はい. 私はすでにそれをやっています

>>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?

>>28329 https://github.com/Panchovix/stable-diffusion-webui-reForge

>>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)

>>28338 whoa nice

I'm back, what did I miss >check /hdg/, picrel Looks like I didn't miss a thing

>>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

>>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

>>28356 no problem btw keep an eye on this https://www.nexusmods.com/bloodborne/mods/45?tab=files

i regret to inform you that wojak is a recognized tag

What's the verdict on vpred v1.0?

>>28360 Been going back and fourth between 0.75 and 1.0. 1.0 is overall a little darker, more contrast, and a little more realistic with some odd changes to lighting. Feels more like a new mix than any sort of improvement but it might complement some styles better.

>>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

merry christmas, everyone

>>28364 holy fucking kino that's the best thing i've seen in a long time

>>28364 Merry christmas

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

Christmas wife

>>28371 Merry Christmas >>28373 Cute wife

>>28364 Nice!

Merry Christmas!

>>28377 Merry Chrismas.

Merry Christmas

>>28379 cute

>>28386 🔩

>>28386 sexo >>28388 christmas brats erotic

oopsie

Is v-pred just harder to train? All my test runs were underbake

>>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

>>28391 >>28392 I'm glad that she's alright.

>>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

>>28411 UOH

pose borrowed from https://gelbooru.com/index.php?page=post&s=view&id=8294248

>>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

that was fun. I even made it into the highlights before getting banned^_^

colorization attempt

Promising

custom udon lora for noob v-pred v1.0 https://mega.nz/folder/LAxGha4I#Jq2eDy7KV2wu2tVuhNjNGg

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

bl-ack!

>>28421 >>28422 Now this is some good shit Thanks

>>28428 Mature and sexy wife

>>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

>>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

I fucking love this model

I'm uohverdosing again.

>>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

>>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

Thank god that I'm going back to work in a few days.

>>28426 catbox for the first one? or artist mix, love that kind of cunny face

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

feral lemma lora for noob v-pred v1.0 https://mega.nz/folder/aRgH3K7J#3988zX24f_nIJzNj9o_Gog

modular psu connectors are the fucking devil

Happy new year, gonna be a extremely busy one.

>>28451 nuked the folder. this krisbox is all that remains

>>28460 works good.

happy new year

>>28466 have a radical new year

>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

>>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

>>28475 cope is cope. i've stopped trying to find reason in it.

personally i find both the "artists are finished" shitposters and actual believers to be based (and that includes myself, obviously)

Year of the snek coming.

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

>>28473 It's a test run, tag recognition is visibly degraded when applied in full strength antifreez3 (tianliang duohe fangdongye) lora for noob v-pred v1.0 https://mega.nz/folder/2JBSkSaR#6yFhJtt-GcWgVN7K_v6AkQ Is there a better tianliang dataset in the wild? This is the one I put up in an hour, password is the url of this board (the portion after "https://", without a trailing slash): https://pixeldrain.com/u/hbR5NkHH

>>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.