>>11184 Requesting Vivian James crushing Anita Sarkessian’s head with her bare hands.

Mixed up a nice watercolor style then made funny & lewd stuff. Index Anime Style - marker [LoRA]: https://civitai.com/models/494434?modelVersionId=556110 CFG Scale Boost [LoRA]: https://civitai.com/models/99619?modelVersionId=947620 Arcade Illustrious [LoRA]: https://civitai.com/models/1270707 (artist:optionaltypo:0.8), retro artstyle, <lora:watercolor:0.9>, (watercolor_\(medium\):1.2), <lora:cfg_scale_boost:0.5>,

Some models newly ported to Pony because I'm ran out of stuff I can do quickly https://civitai.com/models/561054/rosa-after-war-gundam-x-under-the-moonlight-ii-x-under-the-moonlight https://civitai.com/models/906322/james-bond-daily-express-comics >got 2/5 OP images Nice. I'm amazed the Donte one came out as well as it did since I saw the LoRA, threw on a couple of extra tags I wasn't even sure would work, and saw what came out, all in reference to a 12 year video that wasn't even that big (it's "only" 1.2 mil today).

Carol in Hat in Time style. I need to get Demon's Turf and Frogun during the sale tomorrow, then make LoRAs for the main characters. What's the group of Carol, Beebz, Renata, Hat Kid and Nobeta (as seen in >>1058558 ) called anyways?

>>11190 The video models struggle a lot with less realistic art styles, I'm guessing they were trained overwhelmingly on real footage. Or maybe there simply isn't enough animated work to train them well?

hello thread

There's an Illustrious version of Angel's Egg style now. Angel's Egg Style - 1985 anime [LoRA]: https://civitai.com/models/582886 Lines are far cleaner than the Pony LoRA I tried in the past. I dunno if that's a good thing for emulating the style, but it's much sharper / cleaner. >>11189 >Hat in Time style I noticed that style today & downloaded it. Haven't tried it yet. >>11192 hey

I'm Brapping and fapping

How do I tag this hairstyle? Anything better than Tonsure?

>>11195 balding

>>11196 Less than 600 on Danbooru, but I could try it.

>>11197 if you are going by their tags then it should be bald. >The "bald" tag is also used when a character has no hair at the top of their head, but has hair on the sides of it.

Grimphantom style [LoRA]: https://civitai.com/models/182833 Sexy Retro Cartoon 2 a.m. Style [LoRA]: https://civitai.com/models/1292840 Style Prompt: cartoon, flat color, bold lines, <lora:Grimphantom_-_Illustrious:0.4>, grimpcart, flat color, <lora:Sexy_Retro_Cartoon_ILXL:0.2> , src2amxl, cartoon style, Initially, I tried to get a Spumco style by mixing in a LoRA for a female character from Adult Party Cartoon, but it could only do females & contaminates males during hi-res upscale. So I settled on this other toon mix. I voice-cloned CD-i Link, but it's not that convincing, Probably needs an audio filter to sound closer to it. Zelda voice clone is from Age of Calamity.

>>11184 requesting reaction a image based on the /hisparefugio/ OC mascot, "la refugiada"

>>11200 I instantly recognized Link as the CD-I version, so you did well there. It's not spot on, but it was close enough that people can tell. Good message, by the way.

>>11201 not brown and fat enough

3rd of ~12 epoches. This worked way better than expected. Eager to see how it works when I can actually use negative prompts and style LoRAs with it.

>>11201 >>11185 An attempt was made. When the model doesn't recognize the character, the result tends to look a bit more generic.

>>11204 >pic 1 Is that Boondocks style Caius Cosades?

>>11206 No, just "anime screenshot" (though it is Caius). Trying to pull out styles more specific than in Illustrious (or more than the recognized "source_" tags in Pony) that for epoch pics is a crapshoot and doesn't really help with the main point of epoch pics anyways. Caius is largely a test to see if low poly models work for training (it's a 16 to 7 ratio).

>>11201 >>11185 Attempt 2 trying to make her a bit less generic by making her face rounder, giving her fuller bangs and more relaxed eyes.

This could've had more Frazetta influence, but I like it. Frank Frazetta Style: https://civitai.com/models/410093

>>11184 I'm still waiting on that cuties parody of video game shortstacks called "shorties". >>>/ais/5100

>>11209 I wish more game with anime girls had the balls to pull something like this. Hell, even the faces look better and more expressive that some of the shit they churn out.

>>11212 I just know that I enjoy the unworldly feeling that low-poly fan service has and wish there were more fan service games for retro consoles.

>>11210 Viv snu snu

>>11213 Its easier to do a effective art style with a low/medium poly aesthetic than anything uber realistic. I played Atlyss and other than my disdain of the main races the aesthetics made it look better than so much shit that was released over the last few years. Like the goofy particle effects that Dragonball Sparking Zero has makes it look worse sometimes than the much older FighterZ.

>>11207 >>11206 Reminds me of Avatar/Korra, honestly. The South Korean studio that did Korra, specifically.

>>11215 >Simplistic things are easier to generate than more complicated things Go figure.

Do any of you know how many images do you need to make a celeb LoRA with FLUX? I have plenty of pictures but not sure if using all of them would be a good idea.

AFAIK there are no Ren & Stimpy, Spumco or Kricfalusi style LoRAs for Illustrious so I tried to make it happen by mixing in a character LoRA. Soap Girl (Ren & Stimpy Adult Party Cartoon): https://civitai.com/models/1078864 Sexy Retro Cartoon 2 a.m. Style [LoRA]: https://civitai.com/models/1292840 <lora:redhead_girl_Red__Stimpy_Adult_Party_Cartoon:0.6>, cartoon, flat color, bold lines, <lora:Sexy_Retro_Cartoon_ILXL:0.5>, src2amxl, cartoon style, Result was a highly breedable toon Viv.

>>11211 That's a very difficult request and it requires many hours of work.

Anyone recognize this cutie? (I'm surprised it's working as well as it is given the training data)

>>11219 It looks good, but wild fluid looking caricatures are going to be challenging to get right, it's simply the nature of things. Not a bad attempt though, I'd like to see more.

>>11222 ICO posting?

>>11221 Images from final LoRA (comes tomorrow, baring Civitai shitting itself further) + my already published LucasArts 95+96 style. I continue to be amazed at how crap a dataset I can string good results from. I didn't even have to do any real cleaning on this one (just some crops+next nearest size boosting).

>>11220 It's not that hard. There are better ways to do it, but this is the workflow I'm comfortable with and makes me feel like I did something instead of settling with what's given: Generate a base for each character, generate a stage, clean it up with inpaint, composite everything, and upscale. Now, here comes the but. There are no incentives to do this. Stuff made with AI is seen as disposable. So why would I spend time and effort on something for someone else that'll be seen once and then tossed into the bin? I would rather spend this time and effort on stuff for myself.

>>11225 I bet you can't do it.

>>11225 doing them in the low poly style looks cool. >>11226 he already did most of the work thoughbeit.

>What? Yes. I'm Caius Cosades. But, what do you mean, you were told to "Report to Caius Cosades"? What are you talking about? https://civitai.com/models/1350486?modelVersionId=1525410

Guess what artist style this is, I tried it once with a Pony LoRA, but didn't think it looked clean enough. It's still not clean enough with this Illustrious LoRA + artist tag, even after upscaling & inpainting, but still cleaner than it was on Pony.

>>11229 why are they covered in poopie? Duking in dukie; fighting that they do do? poop

>>11225 Damn, that's awesome! I was expecting it in 2D if it were to be done at all but low poly 3d is just as good! Thanks for getting at least this much done!

>>11230 My guess is it's strip mud wrestling.

https://civitai.com/models/1310253/rom-art-style-for-illustrious-v60 https://civitai.com/models/474074/dragon-ball-style https://civitai.com/models/1120415/mount-paozu-dragon-ball-or-illus-background

>>11228 Cool Caius Cosades. >>11210 >Frank Frazetta Viv Very nice.

>>1079850 >cuties

>>1079850 >rumble pack >4 player simultaneous

Here's the fix

>>11238 Now code it into an actual game.

>>11238 So good

https://civitai.com/models/1356307/il-pepper-pumpernickel-peppers-adventures-in-time?modelVersionId=1532105 Pepper Pumpernickel from Sierra adventure game Pepper's Adventures in Time . I have no idea how this worked given I had 19 pictures like pic 4, 6 pictures like pic 5, 2 amateur sketches and the front cover. It does need a style to not suck though.

>>11238 Thanks for the work, I was waiting for a long time for it.

>>11235 Nice. Did you do any manual adjustment on the pants? I didn't really focus on them in training, various styles are all over the place on how it handles them, and it wasn't until I made this LoRA I realized how heavily patched they were.

>>11233 nice

>VERY STRONG >EXTREMELY FEROCIOUS >INCREDIBLY FERTILE

>>11239 And what the gameplay would be? A rhythm game where the character strips for each milestone reached, and if you hit 100%, it unlocks a private dance?

>>11246 Or maybe they strip when you mess up.

>>11246 pull a fast one and make it a kart racer

>>11246 Maybe a stripclub management simulator, like that German game?

Tried to recreate a pose using controlnet.

Made some nihonga/ukiyo-e style stuff with 2b, Kat, Selvaria Bles, Demi-fiend and Zidane Tribal. Some tags that may be of interest: minimalism, ukiyo-e, nihonga And an extra: kanji, japanese text >>11243 No adjustments. I just used the specified trigger words and Better Days took care of it. >wasn't until I made this LoRA I realized how heavily patched they were I never noticed that either until I read the trigger words and saw what was being generated. Kind of feel like playing Morrowind now. Very cozy game.

>>11251 I got an ukiyo-e LoRA, but hadn't played with it much. I only know it generates cool backgrounds.

>>11251 I like how Skynet is bad at understanding pf physics.

>>11241 checked jesus, it being able to do so much with so little is truly nice

>>11251 >Kind of feel like playing Morrowind now. Very cozy game. A mod adding Cyrodil's Kingdom of Anvil released a few months ago and is a great expansion. Entirely possible to play starting in the expansion and never leaving. The chunk of Skyrim and Tamriel Rebuilt are also both good. >>11254 One of the most important pieces of advice I ever got for making LoRAs was someone pointing me to the concept of image "depth" (pic related) and how learning sees things when I asked why my Surface of the Moon LoRA produced such poor results (since I made it from all public domain material and could release the training data). The main takeaway I got is that image quality in the technical sense isn't actually that important as long as edges are clearly defined (which on the moon they are definitely not) and the details aren't obscured. It also explains why patterns and textures in training come out the way they do. Making LoRAs of stuff with minimal data appeals to my sense of perfectionism: When spoiled for training data, I always feel I can do better in the end, but when the pictures I've got are everything (or everything official) of the character, any success is the best possible. Though I do think this one could have been better, even if only marginally so, if I knew the modding tools for Sierra adventure games (I know they exist, but don't know what to use or how) and could just rip her sprite sheet(s) instead of trying to find distinct poses.

https://civitai.com/models/1358239/pd-sarah-anderson-lucasarts-outlaws Another retrain for Pony (haven't bothered posting them. Does anyone care that one of my existing models now has a Pony version?), except this one overhauled the tagging. I should make a v2 for Illustrious with these improvements at some point, though would like to find another image of her first.

>>11186 Seconding this.

>>11251 Reminds me of this Sengoku reinterpretation of Kill la Kill, only this was done by a human.

>>11259 But were they a hero?

>>11257 I don't use a Pony checkpoint anymore, but it's nice to hear you've made progress in your endeavors, anon.

>>11260 que.

>>11251 Janku 2.1, not bad. the colors are pretty good

>>11263 That Raven one looks very nice.

>>11263 Was looking into that update, but saw the same guy released JANKUN on Sunday, a v-pred model that's supposed to have better anatomy. I haven't tried it yet, but I was still having issues with fingers & toes at times on JANKU so I may try JANKUN instead.

Accidentally released two new LoRAs at once again. Here's Sodium Chloride Skeet, also a pair of girls who appeared in one episode and share a base model. https://civitai.com/models/1367824/il-sagebrush-sally-and-elke-elkberg-jimmy-neutron https://civitai.com/models/1368397/il-skeet-employee-of-the-month-jimmy-neutron

>>1078021 This is what you get for non, anon

>>11210 >first image I wish this was really how people dressed IRL. Just cover the genitals and the anus, that's enough.

>>11268 Anon, it's cold. If you want to look at bodies like that, go squat in the mud in Africa or Australia, where they literally never invented anything whatsoever.

>>11268 Obese people aren't nice to look at, anon.

burger. some fries. small shake. maybe some ranch for the chicken nuggets.

>>11271 I like both Ashley and Anby.

>>11269 My fault for not fully explaining. I meant I wish it was socially acceptable to dress this way, not necessarily people dressing this way all the time regardless of the weather or other conditions. >>11270 Most obesity exists because of what they've done to the food supply, so it should rarely even be a thing in the first place.

Likely the strangest LoRA I'll ever do: Animated evil pants. https://civitai.com/models/1371748/il-experimental-when-pants-attack-jimmy-neutron?modelVersionId=1549855

>>11274 Now do Wonder of U.

>>11274 MAKE THEM THICK

this is fine

>>11278 After deleting all my LoRAs for Pony I tried to save drive space relying more on known artist & character tags. But I too have would up wasting a lot of space on LoRAs yet again. Scrolling through civit I often start collecting/downloading them a little too indiscriminately.

Drill hand https://civitai.com/models/1376530/il-drill-hand-weapon?modelVersionId=1555283

>>11271 Burgers? >>11256 >Morrowind Thanks for the recommendations. I haven't played Morrowind in ages, so I'll have to check on some "essential" mods later as well. >Tamriel Rebuilt A friend of mine made a few houses for it a good 20 years ago. It's nice and pretty crazy how the project is still active. >>11263 Pretty nice. Forgot to mention that I also used tags like samurai armor and ninja clothes to change the outfits. Also not including the minimalism tag will naturally give the image more detail.

>>11281 >Burgers? AI-generated images?

>>11271 >>11278 >>11281 I love burgers!

Beebz LoRA. https://civitai.com/models/1382704/il-beebz-demon-turf?modelVersionId=1562428 >>11184 Carol (LoRA for her is linked in Beebz's recomended resources) playing with Beebz. >>11191 With Mini Max, there was clearly 3D models underneath everything. It was clear with how hair and stuff was subject to video game styles physics, and it just failed on stuff outside typical skeletons. With how they do CGI and video game model stuff just fine, it wouldn't surprise me at all if the new, non-web, models did the same thing.

I don't know if WAN etc. have this kind of issue though.

And I accidentally posted 2 models on one day again. Well, here's the G Gundam "This hand of mine" pose. https://civitai.com/models/1385829/il-this-hand-of-mine-shiningburning-finger-g-gundam-g

I got finally apollio to run now but for some reason when I try to use it's TTS prompt there is no audio being outputted and how do I get this program to clone the voice of Renamon?

>>11287 >apollio You mean Applio? From this: https://applio.org/playground -it looks like you can upload a voice sample under the inference section & it'll attempt to replicate that using some existing voice model. But that's based on what little I have skimmed through in the documentation. I don't see any means to tweak the emotion of the speaker so I'm not sure how this works. Maybe just look up videos of someone using it to get an idea.

>>11288 Ah hell I need to somehow find some cleaner audio recording of Renamon, I uploaded a short audio recording of her and it took forever to process and the end result was bad, I am pretty sure the result was bad because there is too much background noise on the original file.

>>11289 There is AI that can remove the background track from audio. IIRC Retrieval-based-Voice-Conversion-WebUI has it in its toolset.

Style LoRA tests with Vivian. https://civitai.com/models/1371624/dirk-erik-style https://civitai.com/models/1375704/birth-days-illustrious https://civitai.com/models/1373599/ccs-chinese-fantasy https://civitai.com/models/1371351/rockos-modern-life-style https://civitai.com/models/1283210/studio-ghibli-anime-style-pony-illustrious

More style LoRA tests with Vivian. https://civitai.com/models/1349354/cartoony-style-or-juukyuu-style-or-flat-colors-or-illustrious-lora https://civitai.com/models/1385017/dustywok-illustrious-style https://civitai.com/models/1361986/illustrious-xl-akamatsu-ken-love-hina-negima-artist-style https://civitai.com/models/1373986/j4gkb-or-shiiros-styles https://civitai.com/models/1324562/style-fuchurberndsoft-illustrious-xlnoobai-xl Didn't do inpainting. Hands are okay. Fingers are maybe slightly less broken on version 2.1 of JANKU checkpoint. >>11263

>>11291 >Rocko's Modern Life Vivian >>11292 >Unteralterbach Vivian Amazing.

>>11292 Juukyuu Vivian needs less detail on the eyes. He always does eyes as solid circles or ovals with no sclera. An exception could be made for making her eyes baggy. Does the style LoRA produce Juukyuu's simple eyes in any other testing you've done?

>>11291 >>11292 AI is getting too powerful, I don't like it

>>11294 It definitely would make the eyes like it should, but I messed them up by forgetting to remove eye-related tags from the prompt while I was switching through the styles. >>11293 >Rocko's Modern Life Vivian I wanted to know what it does with human characters. Decent cartoon look, but seemed to make things blurry. >>11295 AI can even persuade the human using it. The chinese fantasy LoRA really really wanted to include a dragon, shoving all kinds of abstract crap in there, until I was like "fine. just make the dragon" then added (dragon in background) to the prompt so it could draw it. I'm often looking for very cartoony western styles for Illustrious since I had quite a lot of them for Pony. I'm steadily building up my LoRA collection again & wasting a lot of disk space.

>>11290 There is one? Ah thats good well I hope it will work then. Looks like I better have to wait first for my 32GB ram sticks to arrive before I do more AI stuff, because I have only 16GB in total right now and my new ram sticks arrives first at 5th april so its going to take a while before I can finally do anything AI related. Well I can still do AI Chatbots locally with a 7-8B Model so thats not much the issue here but I have shit tons of program running all the time that uses up a lot of ram and I have only 10GB more ram as a swap file on my HDD drive which is slow. Welp I have made some mistakes in my hardware purchase, meh.

>>11296 >blurry Makes sense, given that all the source material has never been recaptured above 480p. It makes it feel vintage (as in: how we used to see it).

V2 of Sarah from Outlaws. https://civitai.com/models/1240241?modelVersionId=1571933 Changing the tags definiately made it more stable, though as pic 3 shows it still sometimes wants to draw her dress lower than it actually is >>11296 Did you try my LucasArts style? I've found it weaker at women than men, boys and girls, but you might be able to get more out of it. https://civitai.com/models/1324713/il-experimental-lucasarts-1995-96-cutscene-style

I'm surprised how well the string >menacing (jojo), ゴ, sound effects works in Illustrious. No LoRA needed: It just works. May be able to strip it to just "menacing (jojo)" at cost of some consistency. ゴゴゴゴゴゴゴゴゴゴゴゴゴゴゴ

Zonos does other languages so I wanted to try generating voices in Japanese. I used a Yukari Tamura voice clone for Vivian's Japanese voice. Male voice is reusing ATHF Carl as I already had a source & it sounded fine in Japanese. I don't know Japanese so it's simple dialogue. Kani means Crab, but sounds like something else. Vivian James [LoRA]: https://civitai.com/models/1228761 Myriad - Illus x Niji [LoRA]: https://civitai.com/models/1226171 Fantasy Western Animation - Phil Bourassa [LoRa]: https://civitai.com/models/1145367 <lora:IL_Vivian_James_Afterlife_Empire:0.7>, vivija, no hairbow, <lora:PhilBourrassaWesternArtstyleNoobLocon_byKonan:0.5>, <lora:myriad_illus_x_niji_IL:0.4>, Myriad, (flat color, anime_screenshot), Although I don't use the small 8-ribbon, the Vivian LoRA helps generate green & purple stripes FAR easier.

Another attempt at menacing Beebz. More ゴ this gen. >>11301 >Carl trying to speak Japanese Heh. How does your Yukari Tamura model do speaking English? >the Vivian LoRA I'm glad it gets use.

https://civitai.com/models/1398320/il-colt-single-action-army-revolver?modelVersionId=1580970 LoRA to make Single Action Army revolver. Temperamental, but works a lot better than asking for a "revolver". >Six Bullets...More than enough to kill anything that moves!

I'm starting to get annoyed by the lack of png support. I guess I'll just match the background to the default css.

>>11304 make one with nips plz

>>11304 In theory you could do webp, but >webp

>>11306 Nope, I tried webp and it gets blocked too.

>>11304 >Blue Gun She's a fan of Lethal Enforcers? Nice.

>>11304 I always did JPEGs for this stuff anyway since hi-res PNGs often have very stupid file sizes especially with film grain or patterns. Sucks we cannot do transparency though. >>11306 >>11307 I've seen transparent WEBMs, but dunno how they're made. Dunno why they cannot enable PNG for boards that are more actively moderated since I thought the spammer was exploiting dead boards where content wouldn't get deleted for a long time. Guess there's also no mechanism to ONLY toggle off APNG support. Spammer will probably switch to using video & get that disabled too.

>>11306 whats wrong with webp?

>>11311 It's a poorly supported new format that a few large companies (Google, Wikia) have been trying to force. Everything it tires to do is done worse than the existing jpegXL, but Google took advantage of owning Chrome to rig the game by removing fully functional jpegxl support to push people to the inferior but Google owned standard.

>>11311 Three reasons >actually worse than jpeg for (subjective) quality to file size ratio the human brain is weird but even though webp is technically on par than jpeg, it actually looks worse >it's poorly coded and actually allowed for an RCE for some time, somehow >NOTHING FUCKING SUPPORTS IT

>>11311 The only problem with webp is that retarded nerds that code things hate it. So theres almost no support for it. Stupid Schizos

>>11313 >>NOTHING FUCKING SUPPORTS IT This site and reddit does :^)

>>11311 Google created it.

>>11189 >What's the group of Carol, Beebz, Renata, Hat Kid and Nobeta (as seen in >>1058558) called anyways? Apparently, fucking cuckchan gave the group the "C-team" name. I don't like it, it just sounds like their games are C-tier at best when I really enjoyed what Onirism had to offer this far. And yes, I know what the C is supposed to stand for, it's still a shit name. Anyways. >>11184 Requesting Carol, Renata and Nobeta casually shittalking American AAA games (or cars, whichever you prefer) in the Top Gear set. The crowd is optional.

>>11317 I never once saw it as "C-tier" until you said so, so it must be a (You) problem.

>>11317 >retard first thought it meant "C-tier" instead of the more obvious Cunny-Team

>>11319 If you had an attention span, you would've read that I knew what the "C" stood for, you blithering idiot.

>>11319 >>11320 Burn.

This is basically proof that all of society is just scripted in Hollywood. https://x.com/WhiteHouse/status/1905332049021415862 https://archive.is/9z5JI

>>11322 Any idea what specific tools (e.g., Illustrious img2img) they used?

>>11323 OpenAI just added a new Studio Ghibli functionality, and apparently everyone's using it.

>>11311 1. poor support, and .png is already superior. 2. most drawing programs won't open the .webp file type. 3. unless you open up the file itself in a drawing program or other and take a closer look, it's impossible to tell whether or not the .webp was saved in a lossy or lossless format; compare that to .png & .gif which you automatically know are both lossless, or .jpg/.jpeg which you automatically know is lossy. 4. as mentioned earlier by the other anon, the quality is worse; the whole selling point is "but look! the file size is like ~25% smaller! now you can host/save more images!" - who gives a shit? this isn't the 1980s. storage space isn't that expensive. as for saving images, the average PC has at least a 1 TB drive, that's plenty of space to store your high-quality ~5 MB .png images downloaded from your favorite booru. if not, you can easily drop $80-$100 on a 4 TB external drive.

>>11323 send a FOIA request

>>11311 >>11312 >>11313 >>11314 Are there any addons for firefox or some greasemonkey script to automatically turn downloaded webp files into jpegs/pngs?

>>11328 Same question, but for Opera/Chrome/WebKit browsers. I'm so goddamned furious when I read a URL that directly ends in JPG or PNG, drag the file to my desktop, and it's magically a WEBP. Everyone at Google needs to be murdered.

>>11324 Do that have a deal with jibbly? I could imagine immediate legal shenanigans otherwise.

>>11325 if not, you can easily drop $80-$100 on a 4 TB external drive. They're a little more expensive than that. Unless you want one that's going to fail really fast.

>>11331 I got mine refurbished, and it's been working fine for me for over 5 years. Although, I don't use it every day.

>>11330 You can't own a style. If you could, Disney would own all anime.

>>11333 But you can own the body of works that AI was trained on, and the law still has yet to fully shake out around that subject.

For me it's the Miquella ones

>>11335 For cuddling only.

>>11334 All the more reason to start exploiting it as much as you can, before the legalfags start regulating your freedom. Worst case scenario we may have to start pirating AI models.

>>11324 It's not just a Studio Ghibli functionality, it's a full blown new model, probably in an auto regression architecture that has insane image cohesion. You can accurately describe an image and it will match your prompt to absurd degrees, and it's especially good in image-to-image cognition, where you can offer an example and it will one shot learn all the features in it, no fine tuning or lora required. Absurd.

>>11338 > auto regression architecture Really? Tons of diferent architectures and they decide to use the oldest, most expensive and ineffective one?, I guess (((Open))) AI has a lot of investors money to burn kek. https://www.youtube.com/watch?v=zc5NTeJbk-k

>>11330 They're shit out of luck, there's no stopping it no matter what they do, the cat is out of the bag, Pandora's box is in smithereens.

>>11339 Being the oldest has nothing to do with necessarily being ineffective. Still, it's not literally just some form of a standalone autogressive image gen model, the image generation capabilities seem to be embedded into ChatGPT just like some multimodal chinese models could also generate images, although I'm unsure if they weren't going for diffusion there either, and I'm unsure if there's more pre-processing happening under the table through multiple ingestion steps either. Case in point, I'm just impressed with the results. I hope we get an open source version of the tech soon enough, and I hope it doesn't need an entire mainframe to run, but I find that highly unlikely. >>1083857 It's $20. I personally wouldn't recommend it. I'm not exactly proud to give OAI money, but I'm an enthusiast in the tech stack and they have, extremely unfortunately, still the biggest lead in AI. Here's hoping the chinese surpass them this year and still keep on the open source business.

>>1083857 Use a (((Google))) Model, you only need an account and to select the proper model. Heres the website. https://aistudio.google.com/welcome/ It can do pretty much the same things that ChatGPT although it has a lower quality and some times hallucinates like crazy. But its free lol

>>11336 Miquella is mine to cuddle

>>11338 Sadly, it's an online only model controlled by Google. I have little interest until local models you can actually own exist. >>11343 It looks like a little more than cuddling is going on there.

>>11343 Make Miquella a girl and it's finally pretty hot AI will help me with that.

https://civitai.com/models/1361986/illustrious-xl-akamatsu-ken-love-hina-negima-artist-style https://civitai.com/models/1115090/illustrious-realism-enhancer https://civitai.com/models/1300857/eye-detail-lora-illustriousflux

>>11335 >>11343 https://files.catbox.moe/lw4iai.jpeg https://files.catbox.moe/2o4vob.jpeg

Please stop posting unspoilered pornography, this is a sfw board and as such any picture posted to this board is shown on tbe front page, this makes this particular board incredibly unsuitable for a porn thread.

>>11346 just inpaint the eyes and details instead. parororo and gogalking eyes look really good. ekrea jan too.

>>11347 Seems like there's a peter under the fabric.

Attempted to make another Photorealistic Vivian. I had more interesting LoRAs for Pony Realism so I still prefer it, but I was able to make the stripes work better on an Illustrious Realism checkpoint. Couldn't get a decent ribbon though so gave up on that. https://civitai.com/models/212737/goddess-of-realism I got this one since it seemed like you could maybe do moody film styles with it. Needed a fairly high CFG to get things where I want them & controlnet helps as well.

>>11351 By contrast, Pony Realism has LoRAs for movie stills. 80s Fantasy Movie: https://civitai.com/models/789313?modelVersionId=1286824 If I use the Vivian LoRA then stripes are easier to get right. Regardless, it seems like any inpainting just makes things smoother/flatter & less realistic so I can't do much there. Would like to try Flux since it's most suited to this, but better upgrade GPU beforehand, otherwise I'll get frustrated with how slow & limited it is on my current hardware.

>>11350 peter is on vaction.

Don't mind me, just coming through

Is there a FOSS version of something like this? https://virbo.wondershare.com/app/talkingphoto/

>>11354 It's getting better and better

>>11357 Wrangling WAN is the new hotness. Just imagine, for $2000 you get to stich together 5 second clips.

>>11358 https://github.com/Wan-Video/Wan2.1 >model requires only 8.19 GB VRAM >can generate a 5-second 480P video on an RTX 4090 in about 4 minutes Sounds like a long wait for me, but at least it's possible. There's a LoRA to make Ghibli animations. Studio Ghibli Style: https://civitai.com/models/1404755 A LoRA to make DOOM gameplay footage. Doom FPS: https://civitai.com/models/1376500 Particle Effects LoRA utilizing the IMG2VID functionality. Particalize I2V Lora: https://civitai.com/models/1360349

>>11359 >8.19 GB VRAM WAN works for AMD?

Png is back. Also a few tweaks to the Ashley Graves pic.

>>11361 cute sip >>1084307 what's going on in this image?

>>11361 Blue Gun was better.

Is this Loss? https://civitai.com/models/1405786?modelVersionId=1589057 | Ɩı || |_

>>11364 lol

>>11364 LOL

>>11348 from your post i gathered the want for a board in which sfw stuff must be spoilered but nsfw can be either way

>>11367 Yeah the public education system is a disgrace, illiteracy is really on the rise. I think they offer GED courses though if you are willing to try.

>>11353 I prefer where he's taller and with funnier eyebrows

>>11369 Nice pits. For me, it's genderswap milf tiddies.

>>11325 >most drawing programs won't open the .webp file type Name one

>>11364 https://civitai.com/models/1405786/il-loss?modelVersionId=1602116

>>11364 >>11372 I figured you might do the other panels, but they'd pretty much require regional prompting to make anything. I also thought you'd put them all into the same LoRA with different trigger tags for each panel, but I guess it's simpler to do separate LoRAs. I've been browsing LoRAs on civit for the WAN model & things are slowly advancing there. I really wanna see more character LoRAs for WAN. Like if we had a Vivian LoRA for WAN model, could generate Vivian videos, but maybe IMG2VID could also do that with images of Vivian. I'm reluctant to mess with WAN until I get a 5090, but only reasonably available ones are in Newegg combos for far over MSRP. BestBuy repeatedly screwed me out of getting an FE. After months of trying it'd be nice to just have a damn 5090 already, but I can't justify blowing hundreds more extra parts that I'd lose money trying to resell. That's money I could use on other upgrades. Stock seems to be improving though since Newegg combos are available for hours now, unlike in the past when they'd be sold in minutes no matter what overpriced bundle of shit they were trying to push.

>>11373 I don't expect the greatest results without advanced tools, but Illustrious knows enough characters it can get decent enough sample images without them. The real question I have though is who should Miku be talking to in the third panel? I don't know Vocaloids well enough to pick one and don't want to add a guy (even if it is accurate to the original) I have a good idea for the final panel's cover image though.

>>11374 Miku just makes me think of the President Scott of Domino's Pizza anymore.

>>11371 Clip Studio Paint & Paint Tool SAI.

>>11370 it's not a man anymore it sucks

https://civitai.com/models/1416146/il-better-boyshort-panties?modelVersionId=1600641 "boyshort panties" is a tag, but it doesn't work very well so I made a LoRA that does it properly. Actually got enough material of other styles for a few LoRAs of other basic pantsu types while searching for this (and one is already trained+scheduled). Also realized how the third panel of Loss is the most abstracted of the four by far. Most of the parodies come down to "two people standing" with no effort to mimic the posing, accessories, or even BG of the original and of the ones that do many only use one or two of the elements. If anyone has more parodies not in pic 5, I'd like to have them. I will try without any more if need be, but I'm confident it won't be pretty (4 is much less abstracted though).

>>11379 Do pic 2 and pic 3 with more obvious cameltoe. Cameltoe is just good culture.

>>11380 All you need to do is add "cameltoe" with whatever weight you want.

Witch from Mercury's Ericht Samaya. How has nobody made a model for WfM's tomboy older sister ace pilot yet?! I just flat out couldn't find a model for her, not even for Pony or SD 1.5. https://civitai.com/models/1419979/il-ericht-eri-samaya-mobile-suit-gundam-the-witch-from-mercury?modelVersionId=1605016

>>1085996 Damn it, I did think it would be...

>>11281 >>11283 (¯﹃¯*) borgers

>>11384 For a moment I thought the text on the hat said MANKO as in まんこ meaning pussy.

>>11384 >11.00 MB

>>11381 Fine, I'll do it myself. I wanted a subtle cameltoe, not like the panties is clearly riding up, but a small detail that is rather apparent what it is if you know what you're looking at. Might be too subtle.

>>11384 >>11388 >>1086173

>>11390 Thanks. You even fixed the "M".

>>11391 >two complete arms Absolutely disgusting.

Which one of you was this? ;^)

>>11393 Probably coco himself

>>11383 For some reason, I completely forgot about her beauty mark. This terrible inaccuracy must be rectified. I also noticed a couple of errors in the book pic. Here are the pics again. I deleted the old post to avoid clutter.

>>11390 thank you

Made some Metroids since he's a pretty cool dude and don't afraid of anything. Prompt for the first 4 is: poster, high contrast, metroid \(creature\), pose, arm cannon, fullbody. Prompt for the last one is: isometric, faux figurine, metroid \(creature\), pose, arm cannon, fullbody. Model is Better Days. >>11364 >>11372 Are you going to make a B^U face lora as well? >>11384 >4608x3584 That's a big burger.

https://civitai.com/models/1421420/il-hipster-panties As a compliment to >>11379 and my longer high cut panties, here's another type of panties for sake of more consistent generation. Got a few more panties in this series since it was easy enough to get the training data for them all at the same time.Next week is going to be Pony retrains (got some more stuff mostly complete though so look forward to them). >>11397 >Are you going to make a B^U face lora as well? No. Someone made a style LoRA for Pony but its been deleted.

>>11398 *next week after Loss 3/4

>>11395 Cute! >spoiler

I'm thinking of making a character LoRA, it's an OC from some guy on discord, does the style of the images of the dataset need to be consistent? Do all images have to be the same style?

>>11401 >does the style of the images of the dataset need to be consistent? Do all images have to be the same style? No. In-fact for certain things its recommended they not be.

>>11402 Realized I should expand this when not tired: When you train a LoRA, you're actually teaching the output patterns and stuff that's consistent between your selection, with some filtering and sorting by tags. If all of a LoRA's training data is of a single style, the result will adopt that style (especially for the character, but the BG too to some degree). For characters this is generally perfectly fine and often preferable if you're doing their native style (indeed, a lot of animated characters are totally unrecognizable outside their native style) and lessen the effect of any style LoRAs applied (essentially it's wearing layers) and especially styles just from tags. This was a bigger issue in SD1.5, but some styles are still pretty extreme and make it much harder/impossible to combine with other stuff. Many of these can be isolated by hitching them to style tags the base model knows ("3D" is a big one. If you make a LoRA using primarily 3D training data and don't tell it that training data is 3D, everything you make with it will be a weird pseudo-3D, but if you say its 3D it works fine. Same for "realistic, photo (medium)", "chibi", "sketch", "pixel art", and a few others). It will really be an issue if you try teaching clothing, poses, etc. using a single source (properly tagged "realistic" aside) because your result will be as much a (bad) style LoRA as it will be one for clothes (etc.). (One caveat is that if a character has warped proportions (e.g., very big head) as part of their style, those tend to carry over and look strange in styles incompatible with such. Pic related)

https://civitai.com/models/1405786/il-loss?modelVersionId=1621579 Part 3 of >>11364

Lady Maria, the Hunter, the Doll and her squid friend showing their mad skills with some insane stunts and crazy tricks.

>>11405 >FromSoftware presents BloodBoard

>>11364 Really like that Dagoth one

>>11378 Do the needful and post the anus sirs.