>>1060208 >Do not lewd Internet Jesus. LET'S ALL LOVE LAIN!!!!1

Nice thread, niggers.

>>5653 I refuse to believe anyone nuts to this

>>5653 Most of those are terrible.

>>5657 You think so? All of them except #2 are really well done gens, better than what I've ever managed to generate. #2 isn't that bad either, but the face needs editing.

>>5651 Requesting horror-themed wallpapers with a PS1/N64/Dreamcast style.

So this is a follow-up from last thread with trying to get any of the UIs loaded on Windows 7. For the full story, you can read them here: >>1057266 >>1057609 >>1057623 >>1057972 >>1058218 So I'll cut right to the chase that these are the issues I'm experiencing: >OSError: [WinError 126] The specified module could not be found. Error loading "C:\Users\(My Profile)\Desktop\sd\venv\Lib\site-packages\torch\lib\c10_cuda.dll" or one of it's dependencies This is the part where I'm largely stuck. Trying to find out where the problem is, I downloaded this program called "Dependency Walker": https://www.dependencywalker.com/ <And plugged in the c10_cuda.dll file, with first image listing where the problem is. And this is the problem that I'm having with Stable Diffusion, reForge, and Easy Diffusion (Though Easy is fussing over a different file, instead it's "cudart64_110.dll"). I did find a source stating that the problem can be solved by downloading the specific DLL files that are "missing/problematic", but it also said that it needs the files to be in the folder of the relevant program or it will cause system-wide issues: https://forum.videohelp.com/threads/412661-TMPGenc-Authoring-Works-7-on-Windows-7 And I do not know where I could "safely" download these files nor where I should put them. And in case anyone is wondering, no, these files having nothing to do with Windows 10 or 11 and me being SOL. These were APIs released with Windows 8 and 8.1: https://archive.md/OxEuY https://archive.md/ggiVI https://archive.md/GWilL So I decided to move on and try other UIs. My experience with Comfy gave my this error: >python.exe - Entry Point Not Found: The procedure entry point PssQuerySnapshit could not be located in the dynamic link library KERNEL32.DLL However unlike SD/reF/ED, I don't know what file I'm looking for in regards modifying it so it will use Cpython instead. The only other UI I've tried is qDiffusion, which refuses to move beyond requiring me to install Pillow and k_Diffusion. Whenever I load it up, the installer requires me to install one of the two plugins, I wait for it to finish, close the installer once it's finished, and re-open the program to find that it now "needs" to install the other plugin. And it never moving beyond this point unless I disable "Enforce versions?", which creates the error that you see in the third image. Until this thread was created, I tried asking some Anons over on PLW for suggestions, and here's some advice that I received: <If you can't get pip, try using Pipx. https://github.com/pypa/pipx <And easy way to change python versions is to use Pyenv. https://github.com/pyenv/pyenv <Here's another thing you can try. Delete the python file in the venv folder of your UI. It actually fixed a problem I had before. I managed to get Pipx installed, but I have zero clue how to use it in relation to any of the UIs.

I tired generating something more cute for the thread.

Another day, another conversion. This time Golden Sun's Sheba. https://civitai.com/models/1133040/il-sheba-golden-sun?modelVersionId=1273810 Does anyone have a model rip for Dark Dawn's Himi (or, failing that, a no Djinn code for DD)? I need some back and side shots of the model to round out a data set so I can make her.

I've found these on a random imageboard, any idea what tool was used to generate them?

>>5651 Requesting milf Vivian James tearfully chiding (You) for (You)r shit taste in vidya.

>>5658 Tried fixing the face up slightly

Another conversion, this time Golden Sun's Karst https://civitai.com/models/1137598/il-karst-golden-sun-the-lost-age?modelVersionId=1279208 Tomorrow will actually be a pair of entirely new LoRAs which will both be crosstrained to PDXL after I've confirmed the Illustrious version works. One for a major anniversary of something I think it reasonably popular here, and an SKS/Type 56 for the Civitai contest (because I had it on the maybe pile and someone did a Remington 870 LoRA confirming Illustrious isn't horrible at making specific guns with training).

>>5651 Can I get the sky Lain vaporizing transexuals like Dr Manhattan in Vietnam?

These are unrefined test images. Was trying different styles for American McGee's Alice.

Ultimately settled on Berserk style for the one I would clean up & refine through inpainting & edits.

>>5661 Fucking FINALLY making some progress. Gave up on trying to figure out how I can use Pipx for the moment. Right now trying to get Comfy to function. So in the run BAT files, I finally figured out to replace ".\python_embeded\python.exe" with the directory to Cpython. Right now in the process of figuring out what special plugins I need to install in order to get things to function, from pyyaml to einops. Right, now the only "non-standard" installs I've had to do was as follows: <pip install torch2.2.0 torchvision0.17.0 torchaudio==2.2.0 <pip install safetensors==0.4.2 Now the parts that I'm stuck on is that if I run the Nvidia BAT, I encounter this error: >AssertionError: Torch not compiled with CUDA enabled But the CPU BAT bypasses that and encounters a different error: >DLL load failed while importing tokenizers Currently waiting on tensorrts to finish installing and see if that fixes the issue with the CPU BAT: <pip install tensorrt-cu12 tensorrt-cu12_bindings tensorrt-cu12_libs The one thing I'm learning from all of this is that the people who made these programs and UIs are absolute retards. Who programs software in such a manner that (1) you have to download everything from the net in order for it to function and (2) never include any of the important files/plug-ins/programs with your software so that it can be loaded on practically anything? Yes, I know that I'm mostly having to build this shit from scratch because I'm using Cpython instead of the pre-baked versions included with any of the UIs, but the ONLY reason why I have to start from scratch is that the files/plug-ins/programs that I need to run this shit are not included with any of the pre-baked versions.

>>5671 >>DLL load failed while importing tokenizers Fixed with: <pip install tokenizers-0.19.1 transformers-4.43.0 And it's working! At the moment, the only "remaining" issues is that my system is declaring that avutil-58.dll, avutil-57.dll, and avutil-56.dll are all missing. <Never mind, another issue Also, it looks like I cannot run the UIs using my GPU as the driver version I have installed is 456.71, and it's not support by NVidia: https://docs.nvidia.com/deploy/cuda-compatibility/index.html#use-the-right-compat-package <Branches R525, R515, R510, R465, R460, R455, R450, R440, R418, R410, R396, R390 are end of life and are not supported targets for compatibility. And I'm hesitant to update as I had fun with AMD drivers years ago, where the latest drivers always, with zero exceptions, caused my CPU to crawl or have other issues.

>>5671 >>5672 Never never mind, it worked. Had to downgrade Numpy to 1.26.4, and downgraded Safetensors to 0.4.1 just in case. Now to see how the model treats Ponydiffusion.

>>5673 put a guide on pastebin for googlers with your issues

The prompt was "anime girl, poofy red hair, smiling, jumping in air" <Six minutes maxing out my RAM <Followed by another 38 minutes actually processing the image I'm going to want a dedicated computer for this, aren't I? Or is there a way to speed up the process without worrying about setting my computer on fire, or seriously damaging the hardware running it at "100%". >>5674 Anon, I honestly don't know WHAT I did half the time, aside from install modules that were listed as "missing". The only exceptions being what I already listed. Also, I don't think this is worth a guide yet when there's still the "missing DLL" issue that could be causing who knows what problems that I don't know about yet. And I want to see if I can install a version of Torch actually compatible with my GPU so that I can offload much of the process.

>>5675 If you could work through the issues to get Forge or ReForge working, generating is faster & they have flags that can be enabled for even more speed. I had an issue where if I used LoRAs I would sometimes, but not always, hit against memory limits causing it to take several years to generate anything. Normally it takes seconds. I'd enabled a few flags like --cuda-malloc & --cuda-stream for more speed, but on startup ReForge also recommended --pin-shared-memory although their readme warns that running OOM (out-of-memory) will lead to a crash when --pin-shared-memory is enabled. I took the risk to see if it helped & IT DID. I have --pin-shared-memory enabled now & I don't run into that issue with using LoRAs anymore. Also note that I never had memory issues on Pony XL, only when I started using an Illustrious XL checkpoint with Illustrious LoRAs did this start happening sometimes. All I had to do was enable one flag to fix it & despite warnings about that flag I haven't had any crashes yet.

Welcome to the rice fields! I made a LoRA to gIve you are waifu a Type 56 (the SKS, not the AK) https://civitai.com/models/1138516

>>5678 I've had enough of you filthy rice-monkies

Doing more tests of artist tags. Just running through a huge danbooru artist list. Last time, a few artist tags had little or no influence while this time every artist I tried as I scrolled the list were recognized. Artist Prompts: (artist:boris_/(noborhys/):0.9), (artist:carnelian), (artist:hews), (artist:iesupa:0.9), (artist:zounose),

Artist Prompts: (artist:hamu_koutarou), (artist:houtengeki:0.9), (artist:m-da_s-tarou), (artist:rariatto_/(ganguri/)), (artist:slugbox),

My two favorite discoveries this time were hews & ganguri. Tried some other characters with ganguri. Cute style. Artist Prompts: (artist:hews), (artist:rariatto_/(ganguri/)),

>>5678 get a job

https://civitai.com/models/1131470 Made Megas LoRA. Unfortunately doesn't really understand the head and only mostly understands the billard ball and eyeball are part of it. Didn't put the original head in training data, but something like it will sometimes show up.

>>5672 >At the moment, the only "remaining" issues is that my system is declaring that avutil-58.dll, avutil-57.dll, and avutil-56.dll are all missing. So it looks like Comfy uses FFmpeg for whatever reason and that's where I'm experiencing the error: https://forums.pcsx2.net/Thread-Loading-avutil-58-dll-failed-and-used-for https://forums.pcsx2.net/Thread-avutil-56-dll However I cannot figure out how to solve the damn issues as I cannot find where to download, let alone place/install, the libraries. I figured I would try updating K-Lite and see if that fixes something but no cigar. >>5675 >And I want to see if I can install a version of Torch actually compatible with my GPU so that I can offload much of the process. After updating my GPU with the drivers released on the manufacturer's website, my GPU's driver version is now 473.04 with a CUDA 11.4.189. Managed to finally find a version of Torch with a working URL, but it's for CUDA 11.3. Currently downloading it to see if it will work with my system.

>>5686 >Managed to finally find a version of Torch with a working URL, but it's for CUDA 11.3. Currently downloading it to see if it will work with my system. It doesn't! This is the versions I was downloading: >pip install torch=/=1.12.1+cu113 torchvision=/=0.13.1+cu113 torchaudio==0.12.1 --extra-index-url https://download.pytorch.org/whl/cu113 And while it did finally allow me run the GPU BAT, the problem was that I couldn't render a single damn thing (From either BAT) as it kept popping up with the error: <CLIPTextEncode "round" "_vml_cpu" not implemented for 'Long' And from what I can tell, only one other person has EVER had this issue: https://github.com/comfyanonymous/ComfyUI/issues/5444 And I cannot find anyone who's actually solved a similar issue beyond just waiting for their software to update. But this does make me wonder if I can use any of the older versions of the UIs. Because Cuda 11.3 was "standard" back in 2022: https://github.com/pytorch/pytorch/issues/75992

Win7 sperg PLEASE

>>5668 >>5652 As I worked on the image it reminded me of lyrics from a song I'd heard. I can clean impurity Wash away with kerosene

>>5689 What model is that? I've not genned anything in about a year

>>5690 Better Days IllustriousXL/PonyXL Merge: https://civitai.com/models/623891

>>5691 Thanks

Seems the model doesn't know Selene from Dragon's Dogma (or just not very well). I tried to get it similar enough, but also had to edit in & inpaint the pinwheel medallion every time. I like the Berserk one as Dragon's Dogma took inspiration from Berserk so it fits well.

>>5670 >>5694 Can't remember if I shared this LoRA that I found. Does Berserk style among others. Got it last week shortly after it updated. I'm using the Illustrious version, but there's also one for Pony. Clean 90s Anime Style (Evangelion, Nadia, Berserk, Dragon Ball, Megaman): https://civitai.com/models/1060869

>>5694 Decent job, recognized her instantly before reading the post.

>>5696 Fug, didn't copy image

Ported my LoRA for the Jacket from Mad Max 2 (AKA The Road Warrior). Has a major fix of the trigger word not breaking things anymore. https://civitai.com/models/1144520?modelVersionId=1287201 Also currently working on porting my Gundam X style LoRA. Any requests (e.g., add some minor character, know a character doesn't work in the Pony version) while I'm tweaking a few things? I've got 42 images to work with before I hit the cap, and a chunk will be spent adding a few more of Pala since she's unstable in PDXL version. https://civitai.com/models/715352/after-war-gundam-x-style-x

>>5696 Yeah, got her looking half-decent. I'd considered mixing her with Re-L Mayer from Ergo Proxy to get the eye makeup closer, but that introduced other influences from Re-L's clothing that made things more difficult, not less. >>5697 She has eyebrows they're just small & thin, but since they're hard to see some people think she's SSJ3. The style LoRA didn't even want to draw eyebrows that small so I had to do edits just to get them smaller. >>5698 I don't have enough good references for Selene to make a LoRA, but my predominantly-Illustrious mixed checkpoint doesn't know her outfit at all. Multiple bracelets, a choker, necklaces: they don't appear. On civit, the only LoRA for Selene was trained for SD1.5 in 2023. I don't think there's a model rip anywhere. My miles-long prompts & photoshop are as close as I can get.

>>5699 https://www.pixiv.net/en/tags/%E3%83%89%E3%83%A9%E3%82%B4%E3%83%B3%E3%82%BA%E3%83%89%E3%82%B0%E3%83%9E%20%E3%82%BB%E3%83%AC%E3%83%8A/artworks?s_mode=s_tag Dragon's Dogma has a camera mode so you can expand on the fan art. Just be sure to use F12 (or print screen I guess) and crop instead of the horribly compressed in-game option to actually take a screenshot, and to tag the images "3D"

>>5700 Oh and concept art.

>>5698 Your Gundam X Lora is my favorite. Danke.

>>5700 >19 images on the entirety of pixiv >despite Dragon's Dogma existing for over a decade There's 24 on Danbooru, but a few are off-model. https://danbooru.donmai.us/posts?page=1&tags=selene_%28dragon%27s_dogma%29 If there were a way to easily cut her out of screenshots it'd be EZPZ. Is there not some extension for Stable Diffusion to identify & separate characters from backgrounds?

>>5703 Pixiv has a limit on how many tags an image can have, so there's actually a lot more under Dragon's Dogma alone but they're totally unsorted. >a few are off-model. Remember you can crop off a picture's head and tag it with "head out of frame" to essentially no issues (aside from fewer head pics in training data). Counting the pictures that came that way (some store listings for replicas) and ones without anyone wearing it (other replica listings), over half the images in >>5698 did it and even in that extreme case the worst that will happen is that it will sometimes default to that as the camera position, but only if the prompt doesn't specify any head details and it's not used as a negative tag (so essentially a non-issue). >Is there not some extension for Stable Diffusion to identify & separate characters from backgrounds? If you're training locally, there's masking. https://civitai.com/articles/8974/training-non-face-altering-loras-full-workflow

i have accumulated a total of 6k buzz on civitai, I can train a flux lora 3 times and 12 regular pony/IL LORAs there's so many things I want to do but I also don't know if I should because I might never finish them

>>5704 I re-installed the game, loaded up a save file from 2016. Save was post-dragon so sky was dark & lighting sucked, too dark. Had one earlier save when Selene SHOULD be in Cassardis, but she isn't since Grigori took her away by then. Had to unpack, edit & repack the later 2016 save file to change the "mIsDarknessWorld" flag to false just to get a normal sky with bright lighting for screenshots. Needed to use ReShade & ENB at the same time because ENB can hide the HUD entirely while ReShade has a ChromaKey thing that can cover the background (mostly, still have to edit out shit). After getting it all together I rediscovered that there is no FreeCam in camera mode. You can only hide your player character & rotate around where they are. It sucks for getting screenshots of anyone but the player. Also if your player character is tall it's difficult to get on eye level with any NPC. Another issue is that ChromaKey doesn't work with hair transparency. Any transparent part of hair has sky, walls or terrain in it that the ChromaKey geometry doesn't hide. I really don't know how usable any of it is, but I have over 50 screenshots. Pretty sure the hair transparency shit ruins it. Whole point of using ReShade ChomaKey was to make it much lower effort to separate character from background, yet there's still bits of background in the hair. Regardless, I don't feel like tagging & training all this since dicking around to get the screenshots was hard enough already.

Gundam X style (Illustrious Version) now out. https://civitai.com/models/1145720?modelVersionId=1288589 Also Mana from Gundam X's obscure sequel manga Under the Moonlight. https://civitai.com/models/1148399/il-mana-after-war-gundam-x-under-the-moonlight-x-under-the-moonlight

>>5705 If you can get it without background, tagging is easy. Just setup the tags for you character in a list head to toe then copy/paste them >>5705 Flux has always been disappointing, and its LoRA creation needs entirely different tags. >there's so many things I want to do but I also don't know if I should because I might never finish them Name some of the things. Some things really aren't that time consuming once you get the hang of it.

>>5707 wtfuck is that dagoth ur?

>>5709 How could you be so naive? https://civitai.com/models/1119721/il-dagoth-ur-the-elder-scrolls-iii-morrowind?modelVersionId=1258420

Made this for that other thread.

>>5711 Did she take off her pants because she lost a match?

>>5712 Naw. Girls don't wear pants around the house, anon. Maybe pajama pants in winter, but some don't even do that. Shirt & panties is their most common (at home) attire unless there are visitors or roommates around. I thought most people were aware of this, but maybe I'm imparting some hidden knowledge.

>>5713 I can imagine that being true but it's more boring.

>>5713 >>5714 **The picture is still inaccurate because she is wearing panties. Real girls don't wear pants or panties. Sometimes a long shirt, which covers the view anyway, just in case, but there wouldn't be a point to taking off the pants but leaving the panties on. They take both off. It's healthy to let the vagina air out as much as possible. Helps prevent things like yeast infections.

>>5711 That looks nice >>5713 >>5714 >>5715 Personally I don't like the idea of my fluids going everywhere so I do at least wear some bottoms while at home but since we are technically still on winter I still wear soft Christmas pajama pants as much as I can but that's because I'm a huge nerd and being naked is a bit icky comfy stuff.

>>5717 Miss, you're aware of the rules. Please follow suit.

>>5711 Damn, AI has gotten really good. If it wasn't for the keyboard being so wank and the cans being inconsistent I would've never known it was AI.

>>5719 Human intervention makes it more convincing, but then that IS human hands fixing AI mistakes. Ideally you get a good roll that needs as little intervention as possible. >>5715 Never knew anyone who does this since panties breathe well enough. Exposing that area to filth & contact against everywhere they sit is less sanitary, not more. If you advocate for free-bleeding next then I'll know you're just trolling. >>5717 Doubt this is a woman, but confirms exactly what I said. >>5713 If you want more hidden knowledge women who live with parents often masturbate with the handle of a hairbrush because the presence of a hairbrush is otherwise innocuous in their bedroom while a sex toy would be potentially embarrassing if anyone found it.

Port of Rosa from Gundam X: Under the Moonlight to Illustrious. Illustrious seems to be pickier about using monochrome data. https://civitai.com/models/1151823/il-rosa-after-war-gundam-x-under-the-moonlight-ii-x-under-the-moonlight?modelVersionId=1295503 >>5651 I'd like to see more of Mana ( >>5707 ) grabbing Rosa.

>>5721 Poking at it some more, the issue may be the Pony checkpoints I used had a VAE by default and Illustrious ones do not not. Using PDXL versions of the Rosa and Mana LoRAs is fine for request.

>>5721 >>5652 Used Regional Prompter & Mistoline, Two character LoRAs (Illustrious), (artist:barbarian_tk) in prompt for art style, Photoshop & Inpainting, Better Days checkpoint

>>5723 Nice.

Made some digimon cyber sluts sleuths. Highly recommend both games. Too bad that Beelstarmon isn't in the game though. >>5708 >Flux has always been disappointing How so? I haven't tried it for fear of cooking my 3060, but I've seen some really cool loras for it.

>>5725 >cook my 3060 >with Flux Anon, I run FLUX on a 6gb 1060 on a laptop, there's 0 reason to think FLUX could run any worse on your GPU

>>5660 >>5652 I can't make a PS1 filter very well so I didn't, but here's something.

Have any of you guys heard of and/or tried InvokeAI: https://github.com/invoke-ai/InvokeAI

My two favorite art styles lately. Rariatto & Berserk.

Port of Gundam Breaker's Misa Satsukino. Tomorrow is a double feature of one port and one experimental original. https://civitai.com/models/1155669?modelVersionId=1299747 Been meaning to ask: Anywhere besides Civitai I should post my models? >>5725 >How so? I haven't tried it for fear of cooking my 3060, but I've seen some really cool loras for it. Among other things >censored by default and bad license >it's overtuned for realistic and crap at any illustrated without a specific LoRA >in-fact Flux is frustratingly inconsistent at anything you don't have a LoRA on (you'll never produce the same cut of panties twice) >"natural language" sounds good till you realize it means you have no idea what Flux actually recognizes (unlike booru tags where it's a clean "highly used tag" "seldom used tag" and "non-tag") >when you do know what it actually responds it, it becomes tags but with extra steps (", the atmosphere is..." to every prompt) >the seemingly higher ceiling of natural language evaporates once you realize that you've got a finite character limit on character count in prompts and "natural language" eats up so much more of it The only things Flux is good at are >realistic >text (and only Latin alphabet: It's just as bad as SD with any other writing system) >making LoRAs with very limited training data, especially imitating very specific styles (I've seen solid ones with single digit training data, while general styles in SD based systems want well over 50+ examples) >>5725 Nice. Can you do Ami looking smug with one hand around Nokia's waist the other on Atsuko's while the two wives look embarrassed but happy for the arrangement?

>>5726 Really? I should give it a try then. How long do you take to gen on average? >>5727 >>5729 Neat Viv and cool Berserk Hex Maniac. >>5730 >Anywhere besides Civitai I should post my models? I only know of tensor.art, but I barely used it and mostly stick to civitai. >Flux That's pretty lame, no wonder everyone is justing using illustrious. Still, there are some really nice loras for it. >Can you do Ami looking smug with one hand around Nokia's waist the other on Atsuko's while the two wives look embarrassed but happy for the arrangement? Remind me who Atsuko is, or do you mean Yuuko? It's getting pretty late here, so I'm not thinking straight anymore. Either way, I'll give it a try when I can.

>>5731 I have no idea why I said Atsuko. Meant Yuuko.

>>5726 Bery nice bug

>>5716 I like the contrast on this one.

>>5730 Realized I forgot to show pics for Misa.

>>5731 >How long do you take to gen on average Around 2:30 / 3 mintues per gen with no upscale (flux already needs a high resolution to gen so I really don't care about that) I use a FLUX merge that requires little steps, 8 at most, with it I have been able to make the most of it.

>>5665 I started to try this, but got hung up on a different idea. Might come back to it later.

Hunyuan video-to-video AI model swaps actor in movie scene to Keanu Reeves. Proof of concept. Needs work, but shows the potential.

>>5738 Seems impressive, though it needs to not fuck with random background details >clock >chairs >pant >lights when looking at >skylight (? May actually be a big sequence of lights) >end of the hallway is deadend instead of right angle turn

>>5738 I might finally watch Last of Us

>>5738 The hair being different (and behaving differently) is the highlight of this. I can tell they chose Severance for the testbed because the backgrounds are all so sterile.

Has anyone gone through the effort of training an AI to produce sprite set/sheets yet? Would be cool to do with something like Doom or Duke 3d just to make more natural templates and variants that could be polished up or thrown into the games without having to frankensprite things together to look "right". Would also be cool to see it done with something like any of the 2D fighting games (Mortal Kombat, Street Fighter, King of Fighters, etc) just to see what sort or crazy shit it would spit out.

>>5742 There's a few. It's doable, it mostly works, but you need an extension to get proper results and its generally not worth it because you'll only ever get a single, fixed, image without concern for how it would function as a multi-posed entity.

Made some Lains. >>5732 I genned some images for your request, but the loras bleed a lot into one another so I still need to pick the better images, fix things and do some editing. I only use this thing on weekends, so don't expect it anytime soon, but I'll get it done. Using Illustrious sure feels a lot more simple than using Pony now. >>5736 That doesn't sound too bad. I thought it would take something like 15 minutes since I kept reading about how resource intensive Flux is compared to XL. >I use a FLUX merge that requires little steps Neat. Can you share a link?

As promised, a double feature First up is a port of my model for making the Asticassia School of Technology uniform for Illustrious. The base model sorta knows this one since it's a trained on a dump a bit newer than PDXL's training is based on, but the number of undistinguished variants in that dataset confuse it and it makes everyone into the show's main character as a side effect. https://civitai.com/models/1156360?modelVersionId=1300571 The more daring is an experiment I made with extremely limited dataset (literally just pics from 20 seconds of footage+one still) because I was curious just how limited a dataset I could scrap by with for a clothing LoRA if flip augmentation was enabled. Came out better than I expected, though I've yet to solve the lighting that piggybacks onto it (if I can find something to add to the data that stops it from bleeding in I'll port it to PDXL). https://civitai.com/models/1153595?modelVersionId=1297564

>>5745 Also if anyone has second and fourth image in the sequence of 7 shown her it would be helpful for further iteration and PDXL version. I, obviously, have never actually played the ero mahjong game the training data is from (there's more ero mahjong games than people outside of Asia who know how to play mahjong) and just got the cutscene+third still off a Booru.

>>5737 In any case, thanks.

>>5665 >>5747 >>5652 I didn't get sidetracked this time & delivered. Art style used (artist:hews:0.7) in prompt. Had to clear some noise from template PNG image because PNG + noise = massive file, especially at 4K resolution. Should be able to layer any game cover underneath the PNG as long as it's close to that aspect ratio.

>>5748 Again, thanks.

>>5748 I like how AI does angry expressions on chicks.

>>5744 >Can you share a link? sure, here's the one I mostly use https://civitai.com/models/645943/flux-unchained-by-scg but there's also this one which is 4 steps (though the outputs are much lower prompt adherence quality) https://civitai.com/models/630820/flux-fusion-v2-4-steps-gguf-nf4-fp8fp16

>>5750 It really depends on model and checkpoint. Sometimes its hideous and you need to play with lowering strength or you get something hideous. It's why smug is my favorite emotion to call: It always comes out near perfect.

An Illustrious retrain of Misaki, from the best card game simulator video game ever made (Yu-Gi-Oh! 5D's World Championship 2011: Over the Nexus). https://civitai.com/models/1163485/il-misaki-yu-gi-oh-5ds-over-the-nexus?modelVersionId=1308826

Used (artist:minaba_hideo) in prompt for art style. >>5750 Depends on model & style. One of my favorites is the Viv in OP: >>5650 Her angry eyes are great & she is pouting her lips slightly with her frown. Liked how expressions came out in this too: >>5723 >>5752 Yeah, smug works well. I probably use (half-closed eyes) too often personally.

>>5685 very cool

>>5738 We'll soon be able to swap in Ryan to claim his rightful role as the black panther

>>5755 I was impressed with how well the body came out, but it just toally doesn't get the head at all. I'd need a bunch (at least 7) screenshots where the head is clearly and plainly in view before I revisit it again.

>>5753 she getting ready to play yugioh or duel masters with that duel disk god damn

Saw a Donte model as was obligated to make this.

https://civitai.com/models/1163570?modelVersionId=1308926 Gotcha Force's Usagi ported to Illustrious. Actually have next 3 LoRAs (other Gotcha Force girls+1 more) already trained and scheduled for post.

>>5730 >"natural language" sounds good till you realize it means you have no idea what Flux actually recognizes (unlike booru tags where it's a clean "highly used tag" "seldom used tag" and "non-tag") >when you do know what it actually responds it, it becomes tags but with extra steps (", the atmosphere is..." to every prompt) >the seemingly higher ceiling of natural language evaporates once you realize that you've got a finite character limit on character count in prompts and "natural language" eats up so much more of it Damn this is terrible. I really like the realistic stuff it can do, but this makes me not want to use it.

Is there a booru for AI furshit that doesn't have completely retarded moderation (e6ai)?

>>5762 Renamom! :3

>>5651 Anton Chigurh (No country for old men) having a firefight with Judge Holden of Texas (Blood Meridian) and Jeffrey Epstein. Good Luck.

>>5760 https://civitai.com/models/1163680/il-mana-umihara-gotcha-force?modelVersionId=1309051 Mana. Colors fucked up due to Civitai issue but the LoRA (and one for Usagi) are fine.

>>5762 The only other AI booru I know of is aibooru.onllne, but their focus is on anime. You can still post furslop there.

What would you consider best practice when embedding metadata in files that you've inpainted, upscaled, or otherwise modified? Would it be better to copy the original prompt that made the base image onto the final edit, or just keep whatever your final inpaint/upscale embeds without editing it? I'm leaning toward copying the base image's metadata onto the final result so that it's at least theoretically possible for someone else to generate the unedited version, since just dumping the parameters attached to an inpainted upscale into your generator without the edited init image would give something wildly different, especially if the seed was changed. In the end it doesn't really matter but it's triggering my autism and I'd appreciate some thoughts on it

Forgot to mention Orochi/Rin is up https://civitai.com/models/1163780?modelVersionId=1309162 >>5651 Usagi, Mana, and Rin playing some kind of game together.

Illustrious port for Karin. I need to replay the GCN game sometime to get TD for Marcia. https://civitai.com/models/1163907?modelVersionId=1309308

>>5770 Oh deer.

>>5770 Finally a Disney live action remake that looks worthwhile

Ported Alma. Hail Zeon! https://civitai.com/models/1177925/il-alma-stirner-gundam-code-fairy-code-fairy?modelVersionId=1325481

>>5730 Ami, Nokia and Yuuko sharing some friendly hugs. Note that none of these gens actually have Ami with her hand on Nokia's or Yuuko's waist, because those were a total mess and required some heavy editing to look decent so it wasn't really worth it. Unfortunately the loras collide and bleed a lot which ends up messing up their clothes, colors and a bunch of other things most of the time. Hope you like these. >>5751 Thanks. I'll have to check these out later. Might as well share some Flux loras I'd like to try: https://civitai.com/models/709157/deus-ex-style-flux (Deus Ex) https://civitai.com/models/768707/silent-hill-ps1-in-game-style (Silent Hill) https://civitai.com/models/765791/jet-set-radio-style-flux (Jet Set Radio)

>>5775 Nice.

>>5773 Wow, that's great. Could you do one where she's completely naked?

>>5775 Damn nice. The cyber sluts need more art. Specially Rina and her delicious spats. >>5775 What do you guys think about AI and the future of manga? https://inv.nadeko.net/watch?v=8D4kNb81VIE https://inv.nadeko.net/watch?v=FgQNZNCoaHo

https://civitai.com/models/1182979?modelVersionId=1331385 Mia port as part of Code Fairy series >>5651 Alma and Mia in Gundam X style doing yuri things.

>>5777 >>5652 Alright.

>>5651 Requesting huge tittied Vivian lactating large amounts of milk. Similar to pic related

>>5651 >>5781 Seconding this but unfortunately expanding to include any weird shit; aliens, mecha, those ten-arm boss monsters from Bloodborne. The milk must flow.

>>5782 >those ten-arm boss monsters from Bloodborne. Amygdala or the Wet Nurse?

Because you know more about this stuff. What's the deal with the new chinese AI and Nvidia?

>>5783 both >>5784 From what I understand, investors were very clueless about how LLMs worked and now realize local LLMs exist. None of this is new, people have been doing this with Tavern and Kobold for ages, but techbros had no idea what those are. It's a panic sell as they finally get you don't need insane hardware to train and run a chatbot.

>>5785 What about images and video? The only thing I'm wary off is about it being chinese.

>>5786 It is censored to a degree, if you ask it about Taiwan it'll get all jingoistic on you. However, you can run it locally and offline. As far I'm aware the local versions can't "phone home", the models are also open source if you're really worried. I don't know a lot about their multimodal model Janus-Pro, it's very new. I think it launched just yesterday. Here are some images from it, they look nice enough.

>>5782 See >>5659

>>5651 Requesting Dreamcast/Magazine render of Mark

So, how can you make the same character use different poses? do you need a trained model for each character? is there a tip to use references? Using Krita with stable diffusion.

>>5790 Trained LoRA for character (unless model already knows character) MistoLine ControlNet for poses (use drawings or photos for poses) Regional Prompter for multiple characters (different prompts for areas of an image)

>>5787 How's the image quality of this new chinese AI compare to other AI?

>>5791 I see. I just wanted to make one character reference for multiple poses for TTRPG purposes.

>>5792 I'd argue it's worse in some ways, but apparently the big deal is how performant it is. It's based on DeepSeek-R1 which is reportedly much lighter.

>>5790 >>5793 Getting an original character to appear consistent between generations is kinda a mess. You need to either have the character built in (only the most popular characters like Miku, Touhous and Evas will count for this) or train a character LoRA. Both of which are non-viable for making an original character, and that's even before getting into contamination from other LoRAs. Here's one creator's (fairly detailed) prompt with five different style LoRAs (the contamination is particularly bad here due to the brute force method of style creation the author uses, but the point still stands). Consistent posing is pretty easy with ControlNet. You can freely pose the resulting characters with some skeletons. >is there a tip to use references? As in reference images? That's image to image or (if you've got 20+ consistent but non-identical ones) training a LoRA.

>>5795 Does this video explain ControlNet well? https://www.youtube.com/watch?v=0vkZ9wvcZag

>>5795 I was trying to make different poses for this character, but instead I can just lightly change his clothes. I guess I'll have to redo the promts and edit the result each time I want to do something new.

>>5781 >>5652 Somewhat taxing on Vivian. Regardless, the A.I. genie granted your wish for excessive gamer girl milk.

>>5798 It's even better than I thought it would be.

>>5659 So THAT's what happens when you get kidnapped by a cave monster in dwarf fortress

>>5795 >Original Pokémon >Fairly OddParents >Winx (why the fuck do I even remember the name of the show, never mind the art style; I never saw it) >no clue at all >Toriyama It's amazing what these models have been able to accomplish in just the last 10 years by stealing everyone's data online.

More redheads. (artist:hews) in prompt + CFG Scale Boost: https://civitai.com/models/99619?modelVersionId=947620 Model knows Lina, but doesn't have knowledge to generate her accessories. Wants to make pauldrons for example, but fails every time so I have to put that in the negative prompt. It's good enough to gen her in other outfits without using a LoRA though.

>>5801 Those models are literally just mass screencaps of the original source material, not online pictures.

>>5803 >mass screencaps >not online pictures. Your point is correct, but I think you're expressing it wrong. And besides, he said stealing data, not the art itself, so he's not exactly claiming the AI is plagiarizing in the same way most retards do. The AI does indeed use a ton of data.

Other two Code Fairy girls ported. https://civitai.com/models/1185789/il-helena-hegel-gundam-code-fairy-code-fairy https://civitai.com/models/1188822?modelVersionId=1338294 >>5804 >>5801 The base model is almost certainly covered under fair use. Research and teaching are, explicitly, exempted and it never says it has to be teaching humans. It's a clear yes on the amount/substantiality test (nothing is stored or reproduced and it would be physically impossible for it to be otherwise, since it would require cramming hundreds of millions of images into less than 10 GB), and the effect of the use upon the potential market for or value of the copyrighted work ("the market is more crowded" is not a real impact for this, otherwise nothing would pass it) and even arguable on the nature of the work when related to what the model actually uses (the fundamental characteristics of a style or object is non-fiction, especially in the mass of angles and shapes models actually use). Hyper specific LoRAs trained on a single source or copyright protected subject would have a harder time (I release all of mine as non-commercial novelties)

Forgot pics.

>>5801 >>5803 >>5805 I'd like to note the copyright status on most AI models is unclear, but likely very illegal. Researchers seemed to know this to some extent because many training databases didn't directly give you the images, they gave you millions of hyperlinks to download yourself with a batch file. There are a lot of lawsuits going on, mostly involving OpenAI, and I don't foresee them going in their favor. Understand that regardless of your personal feelings or views on the law, courts usually side with IP holders. American courts especially. Copyright and IP are dogmatically protected.

>>5801 >no clue at all Looks like Bleach but I can't place where on Kubo's style development it actually stands.

>>5807 No, they don't give you the images because distributing images in whole you don't own is definitely illegal, but linking to publicly hosted images on the internet isn't. >There are a lot of lawsuits going on, mostly involving OpenAI, and I don't foresee them going in their favor The one lawsuit over this I'm aware of was tossed out because the people bringing the suit had no idea what they were talking about and the judge, surprisingly, had the basic technical competence to understand that the size of the models is insufficient to actually be distributing the data

>>5807 >>5809 Whether the training of AI models is made illegal or not, the output of AI models (especially existing ones) will be copyrightable in some manner in a way that benefits corps. It just saves too much money for it to be outright made as a derivative of the scanned works and thus copyright infringement. I don't think its likely the models themselves will be made illegal for similar reasons. Too much money for, not enough against. Is there even (ongoing) cases regarding the models themselves?

>>5810 What everyone misses over the one ruling is it wasn't about the copyright eligibility of AI images, it was about if the computer could hold the copyright, which is a straightforward "no" in law.

Coming right on the heels of West Taiwan destroying the AI market: https://archive.ph/w3SqL >Dreamcast's Ad Campaign Finally Comes True In 2025 – The Console Really Is "Thinking" <This was, of course, complete marketing nonsense, but hackers have actually made it more convincing by porting over the llama2.c 'Large Language Model' to the console. <"Thanks to sauce's llama2.c LLM port, your Sega Dreamcast can now become an AI-driven sentient being which can generate responses to user prompts," says @falco_girgis. "As it turns out, the SH4 can actually do a decent amount to HW accelerate the core matrix routines, despite them being a bajillion-by-a-bajillion elements, rather than the typical 4x4 affine matrices seen in computer graphics."

Ruins of Angkor Wat. Combine it with random background stuff to get a fantasy asthetic https://civitai.com/models/1189605?modelVersionId=1339254 >>5651 A smiling Akari slaps Kaori's ass. Kaori looks back at her surprised and horrified. Outdoors on a snowy, overcast, mountain. https://civitai.com/models/177434/akari-hayami-wave-race1080 https://civitai.com/models/1197088?modelVersionId=1347902

What AI do you recomend to make video from images? If possible something you don't have to sign in nor install but I'm already asking too much.

>>5814 video is way too early into it's lifecycle to be like img2img or txt2img, the only way to get exactly what are you asking for are hugginface spaces but running video AI models is pretty expensive and resource consuming so you might be able to generate 1 video before they ask you to create an account. https://huggingface.co/spaces/RageshAntony/HunyuanVideo https://huggingface.co/spaces/THUDM/CogVideoX-5B-Space https://huggingface.co/spaces/Lightricks/LTX-Video-Playground they also might have some restrictions, take that into consideration if you wanted to make nsfw stuff. hailuoai is apparently the best but you need an account and pay for priority.

>>5815 >77.58 KB 1024x1024 image.webp) I would motorboat the top right off those knockers. Come to Daddy miss Luigi.

>>5815 I guess I'll have to wait more to animate my stuff, or fall into the trap and sign up into some AI video service.

DeepSeek-R1 is now live and open source, can I set DeepSeek-R1 up locally? What guide should I follow to set DeepSeek-R1 up locally?

>>5818 What's DeepSeek-R1?

>>5819 DeepSeek is an LLM (Large Language Model) used for chatbots. It's supposedly revolutionary because it compartmentalizes & distills things in a way that makes it faster & more lightweight than traditional LLMs. >>5818 Personally, I'll wait until it's uncensored before I bother with it.

Deepseek also released an image model called "Janus" haven't really tried it locally nor seen any results but it's apparently better than SDXL and DALL-E 3, have you guys tried Janus? >>5818 sure, there's distillations for it to be able to run on something as little as a laptop with 6- 8vram not sure how that works as I have never been interested on text AI. >>5820 so anti-china stuff is hard trained into it? >>5817 maybe it's for the best, as of now the video part of this tech is still pretty resource hungry and it hasn't had it's own "Stable Diffusion" moment yet.

>>5821 deepseek is open source so any kind of rails can technically be removed if you run it locally, I think the same applies to janus I tried janus on huggingface and I wouldn't say it's on par with dall-e3, perhaps only by virtue of not having as extensive a data set of scraped images as dall-e3

https://civitai.com/models/1193910/il-toris-summon-night-2-2 Toris from Summon Night 2. Going to be doing Summon Night stuff till the 8th, so if you've got a problem with that, just take it easy! >>5651 Toris (LoRA linked in Toris's recommended resources. Both have Pony versions if you prefer) making a pinky swear with Hasaha on a moonlit night.

>>5823 She looks good. Japanese school girl, but a little futuristic without going into cyberpunk territory.

Gwendolyn from Vanillaware's masterpiece Odin Sphere. >>5778 >Rina A man of culture and refined tastes.

https://civitai.com/models/1194182/il-amer-summon-night-2-2 Another girl from Toris's game. >>5824 She's actually from a fantasy series. It has limited packets of higher tech around due to the horrific mess that (often one way) summoning magic causes in the series (the first game in the series is actually isekai from before that had a name).

>>5748 What do you use to clean noise? It's becoming a pain in the ass with inpaiting that the images get noisier and noisier.

>>5827 Surface Blur in Photoshop can clean off general noise, but can also wipe out details that may need to be added back in using a mask of the original image. It's a quick way of doing it that isn't perfect & may even cause lines to lose smoothness, but for a 2D style image it's a relatively low-effort fix. I've had the opposite problem with inpainting noise when I use a LoRA that introduces noise as part of the art style. In those cases, if you inpaint anything on the upscaled image, that area winds up with finer noise grain & you have inpaint everything else or it won't match.

I've gone and converted+scheduled my next 7 days of LoRA (max Civitai will accept scheduled posts for), which means I'm actually in site of finishing (last conversion should up up on 18th barring some new setback/obstacle). Decided to clean a dataset I had laying around for my next original (I've actually got a list of 28 things I'd like to try. Some I've actually got preliminary and uncleaned, untagged datasets for, some of which are just vague maybes). Anyone have any non-synthetic Vivans with well drawn ribbons that aren't in pic-related? I'm going to support both clover and ribbon versions.

>>5829 There are many Vivians on our drawfag booru: https://ourobooru.booru.org/index.php?page=post&s=list&tags=vivian_james The ribbon is drawn in various ways so I'm not sure how you narrow that down to a consistent shape.

Made some Bloodbornes.

>>5831 And some Marias.

Another Summon Night girl converted to Illustrious. https://civitai.com/models/1202765 >>5830 Got mine from there. Got a few more on a double check though. >>5651 Requesting Frank Castle https://civitai.com/models/1041262/frank-castle-the-punisher-2005-from-video-game-lora?modelVersionId=1360249 giving Katie Power https://civitai.com/models/1115466/il-katherine-katie-margaret-power-energizer-power-pack-earth-5631 a nice piggyback ride

>>5826 Neat, I might be confusing it with the Summon Sword series maybe.

>>5834 Summon Night: Swordcraft Story is a spinoff series. Next 6 are from it. Would be 7 if I could get more pics of the girl on the right.

>>5833 Why does even the ✡Fandom✡ wiki for this franchise have only a single image reference to this character and literally nothing else? I love (her?) outfit. The colors are great and the design speaks to me. We need more herringbone dresses in the world.

>>5836 Given how Summon Night 3 was never released outside of Japanese, I'm surprised there's anything at all.

>>5832 >>5831 Those are good anon even if i like women to be the ones being carried instead

>>5831 Sorry for double post but what prompt you usedf for the first pic? That lovecraftian design is amazing

Another redhead, but also just brown girls I guess. Prompt tags for the style: (artist:barbarian_tk:0.9), (artist:mika_pikazo:0.6), <lora:cfg_scale_boost:0.7>, CFG Scale Boost linked here already: >>5802

>>5838 >>5839 Glad you like them. >i like women to be the ones being carried instead Well, that was my original intention, but instead I got the doll carrying the hunter and looking pretty proud of it as well. It gave me a good laugh, so I had to post that one. Nothing a quick prompt can't fix though. >first pic Prompt: movie poster, sharpie art, simple background, eye background, eldritch, abstract, horror, tentacles, bloodborne, bloodborne hunter, tricorne, pants, standing, from below, saw cleaver, flintlock pistol Negative prompt: deformed, mutated, ugly, disfigured, long body, lowres, bad anatomy, (bad hands, missing fingers, extra digit, fewer digits:1.2), cropped, very displeasing, (worst quality, bad quality:1.2), bad anatomy, sketch, jpeg artifacts, signature, watermark, username, simple background, And a catbox of the upscaled version with the full metadata: https://files.catbox.moe/gpt91u.jpg The model is "Better Days": https://civitai.com/models/623891 Note that I had to edit the text manually a bit. Everything else was left unchanged though.

First of many Swordcraft Story models, featuring the first game's heroine Pratty (who kisses other girls all the time). https://civitai.com/models/1202823

>>1067351 >>5840 Will there be an uncensored version of Ed?

>>5843 Naw, I'm done with that one. I generated plenty of Ed lewds when I was still using a Pony checkpoint: >>>/ais/5221 >>>/ais/5223 >>>/ais/5251 >>>/ais/4953 >>>/ais/4983

>>5831 >>5841 Tried this on Better Days checkpoint: (sharpie art, hatching /(texture/):1.4), Looks neat.

Do you like ninjas?

People talking about brown tomboys? Here's Swordcraft Story 's Razzy https://civitai.com/models/1203339

>>5846 Can you do them naked?

>>5848 But anon, how will you know they're ninjas if they're naked?

>>5849 They can be holding ninja weapons. They can wear throwing stars on their nipples, and hold sais between their buttcheeks.

>>5849 Isn't there a anime that some ninja girl kill someone with her pussy? I forgot the the name

>>5851 Ninja Scroll had a lady ninja who's body was poison so if you fucked her it would kill you.

Today's LoRA is Swordcraft Story's cute tomboyish, siscon, onee-sama Sanary, https://civitai.com/models/1206841?modelVersionId=1359102 Non-Swordcraft Story LoRAs resume Saturday (or early Sunday on other side of Atlantic)

>>5650 My PC is shit so i'm thinking of getting a cloud instance to generate stuff, do anons recommend any of the sites in the rentry in specific?

>>5854 for now your best option is using a website like Tensor.art or Civitai, both are credit based but they are pretty easy to get, here's my recommendation if you want to get the most out of your credits, >Use TensorArt: https://tensor.art?source_id=njexo13lkke1pvAuan318BEk >You get 50 credits that refresh each day at 5am >Select the checkpoint FLUX Fusion V2 >set steps to 4 as well as CFG to 4 >average prompt will cost 0.4 credits no matter the size of the image Use website on PC as it's the most intuitive version of the tool. Clould stuff has been a bit of a complicated thing since 2023, google realized a bunch of people were using the generous free tier of colab for multiple AI models that included webUIs so they limited that, now you have to pay for units and since then the cloud Stable Diffusion scene pretty much died

>>5855 I asked because i wanted to use this one >>5841 but seems like there is no option to do it on Civitai so i might just try something else instead.

>>5856 pretty weird, I have seen completely new models being usable on the online generator day 1 on civitai and this is a pretty popular anime model. I guess you could try downloading the model and then hosting it on Tensor, a bit unethical and not sure how long it would take the checkpoint to be ready on Tensor as I have seen newly added models take from a few hours to a few days but If you really want to try out that model without spending any money, it's probably the best way to it.

>>5857 I wouldn't mind spending a bit of money if necessary, but a few days is not a big deal, i can just test others while i wait.

>>5857 Civitai allows any LoRA (or variant) of supported base models to be used for on-site generation, but only subscribers can enable checkpoints (probably because they're comparatively huge and relatively easy to hide bad stuff in). >>5855 I'm interested in knowing other model hosts for my many models if anyone has any requests. Won't actively use them, I don't want to have a single point of failure for distribution (and while I enabled permission to use on other sites to avoid any Planet Elder Scrolls/Neverwinter situation, I might as well be the one to earn any credits my models make).

https://civitai.com/models/1207025/il-rifmonica-rif-summon-night-swordcraft-story-3 Swordcraft Story LoRAs are over. Now Swordcraft Story 3 LoRAs, starting with the heroine Rifmonica (Rif). Translation never ever.

>>5651 Slightly different request to a lot of the ones here. 3DPD warning. Try img2img-ing these into other characters or species. I have no absolute suggestions, but I was thinking Xenomorphs, robutts, Draenei, Warframes, maybe R63'd Doom demons. My only real request is to not go for random anime babes, no disrespect but there's so much generic AI anime art and I'd enjoy seeing a different flavor of slop. I had Pyro in mind for #3 I've tried experimenting with im2img myself in the past, but I've never got it to produce what I want. But that was over a year ago and a lot of you have more experience. Even if they come out fucked up, I'd like to see the result. I like #1 so much goddamn

>>5861 I want night elfs and bloof elfs as long there is a change in the face style i find them ugly most of the time >Second pic If face was better it would be quite good

>>5862 >If face was better it would be quite good I'm amused how you're talking about her like she's a drawing. Assuming it's tweaked right, img2img shouldn't copy the face exactly or even at all.

>>5863 It's just that i don't like it,there isn't a deeper meaning in that comment.

Murno, the kiss stealing fivehead gir froml Swordcraft Story 3 https://civitai.com/models/1207050/il-murno-summon-night-swordcraft-story-3?modelVersionId=1359365

>>5865 What's the style of that 3rd one?

>>5865 Big foreheads are a gift of God

Last Swordcraft Story girl (at least till I get more pics of Lynn), the Summon Beast Rufeel. https://civitai.com/models/1208231 Also a preview of coming attractions...

>>5866 Gundam X >>5707 I put way too much work into it (nearly 1k images, all hand tagged), so I show it off in most of my other LoRA galleries. Also Gundam X is a good series (it's even officially on Tubi for free now, so if you're in NA you don't have an excuse to not watch it)

>>5869 It has been free to watch downloaded from irc for several decades as well you know.

>>5870 Yes, but now even normalfags don't have an excuse to not watch it.

>>5652 >>1068905 >>1068971 >Porn when? Now. Some variation, but got the general look. Sketchy Valentine Style (used at 0.8 weight): https://civitai.com/models/1209865 >>5868 >>5869 Sometimes they "just work" while other times the stripes on Vivian's hoodie are a pain in the ass to get looking right. Didn't know if an Illustirous LoRA + checkpoint would be any better at that than the Pony one was. Same issue getting stripes to land consistently in the same place on socks. The AI just draws stripes however it wants.

>>5840 More Edward W. W. Denslow: https://civitai.com/models/1183830 Vixon's Illustrious Styles - Oil Painting: https://civitai.com/models/1112934

I tried to make some reaction pics.