>>6120 You can make pixel art use a consistent pixel grid with the pixelization extension: https://github.com/filipemeneses/comfy_pixelization

>>6121 Thanks, but sadly, I can't use it. I'm using a very limited and very weird workflow. I managed to run the pixelization directly using colab, but right now, it's too wonky, so you guys will have to endure this faux-pixelization. The best I can do is use nQuantCpp to mess with color quantization and dithering

>>6118 >2nd pic >a reshiram that's basically autistic about taking meaning at face value huh...

Had this audio for a while, but didn't make any image for it. Used torikun LoRA again. Forgot to share link earlier, this is the same: >>6101 Redtvpe's Illustrious/NoobAI Styles - torikun [LoRA]: https://civitai.com/models/1009396?modelVersionId=1131455

Fantasy Western Animation - Phil Bourassa NoobAI [LoRA]: https://civitai.com/models/1145367 Prompt for style: <lora:PhilBourrassaWesternArtstyleNoobLocon_byKonan:0.8>, flat colors, anime coloring, anime_screencap, (artist:optionaltypo:0.1), Decent "cel-shaded" like look with this.

Forgot to shill my latest LoRA: Aile and Prairie from Mega Man ZX. I'm surprised prompting both of them at once mostly works. https://civitai.com/models/1318782/il-aile-huxzx-and-prairie-mega-man-zx?modelVersionId=1488879 >>5651 These two being lesbians.

>>1078038 >clang <just a human (but she's totally a robot bro) READ IT

>>6127 That's Roll Caskett from the Legends games, not robot Roll from mainline games. The joke is that she really likes machines.

>>6128 Ah, so she's the clang enjoyer? Good woman. Those are well done by the way I'm just being bitchy. But why are they so blown out? It's obviously intentional, but why?

>>6129 Maybe because I'm mixing two style loras that both have strong contrast and colors. 1990sPCstyle and DCAUstyle. They didn't seem blown out to me while working on them, but I can try reducing saturation and contrast later.

>>1078038 >>1078053 You can sometimes recover blown out skin with a curve unless it's pure white, then you're stuck doing more involved edits. How do I gen squinting eyes? I've tried squint & squinting in prompt, but it never seems to affect the bottom eyelid for some reason.

Used character LoRAs to make it easier to get some details right. Vivian IL [LoRA]: https://civitai.com/models/1228761 Lain IL [LoRA]: https://civitai.com/models/328752 Used this again since it can give a nice animation-screencap look. Fantasy Western Animation - Phil Bourassa NoobAI [LoRA]: https://civitai.com/models/1145367

https://civitai.com/models/1317888/pd-amy-rose-outfit?modelVersionId=1487822 It's >>5992 but for Pony >>6130 >can try reducing saturation and contrast later. I remembered this series existed. No idea if they'd help with what you're doing or if you already know of it. https://civitai.com/models/1093089/il-slider-or-tweaker-color-temperature-saturation-brightness-or-ilnoob

>>6131 Oh that seems easy enough. I'll try it with the other pics. >How do I gen squinting eyes? I've tried squint & squinting in prompt, but it never seems to affect the bottom eyelid for some reason. That's the Anya meme face, you can get it with the tag anya's heh face \(meme\) You can lower the strength to get less squinting. If you're using a1111 or Forge I also recommend installing the tagcomplete extension, it autocompletes booru tags and it helps a lot in finding highly specific tags like that.

What gpooyou are you using to render your shit?

>>6134 That's a neat tag. I'll have to remember it. I'd wouldn't expect an autocomplete to help find that one. Even the Danbooru wiki doesn't list it as a facial expression. I will use this as a chance to shill the effectiveness of "smug" as a facial expression though: Lots of other expressions will be insufficient or so hyper exaggerated they ruin a gen, but smug has never failed me. Incidentally, I like going through booru tags with 1k+ uses in lowest frequency to highest. Finds you all kinds of weird stuff worth trying.

Alright, I managed to restore the skin tones. I also deleted the old posts so they don't clutter the thread with repeat pics. >>6135 RTX 4080 for me.

>>6136 > I'd wouldn't expect an autocomplete to help find that one. That's exactly how I found it though. It autosuggests multiple tags as you're typing, and that one got suggested when typing Anya.

>>6138 Oh you can definitely find some weird tags through auto-complete, but I meant it wouldn't help someone find that was the tag used in a gen. I just found "kubrick stare" is a recognized tag.

>>6135 3080 Ti Can't get 5090 for muh AI since Bestbuy & Newegg suck. Got a kit of 32GB RAM recently since SD kept choking on 16GB. Read about the issue in comments on civit.ai Figured video memory was the bottleneck, but it was system memory. When I'd first add a LoRA to prompt it'd generate significantly slower until I interrupted & tried again, then it'd be fine & could even add or swap LoRA. On 32GB RAM it doesn't happen anymore & I can run other programs without worrying about RAM usage. May also benefit video editing, dunno. >>6137 Looks better now. >>6136 >tag autocomplete Sounds very useful. I had something like that in the early days during first Novel AI leak. Can't recall what the WebUI was called, but it wasn't A111 although that existed. I think there was a JSON file with tons of Danbooru tags in it. My first NAI gens are dated October ‎13th ‎2022.

>>6140 https://gpu.userbenchmark.com/Compare/Nvidia-RTX-3080-Ti-vs-AMD-RX-9070/4115vsm2397468 If the 9070 had 20gb, I would snag it, now I'm conflicted between it and a 20gb 3080ti. They are both the same price, but I haven't seen an AI comparison yet. https://e.tb.cn/h.TM0941U21JnRkMa

Retro Anisthetic IL [LoRA]: https://civitai.com/models/1287151 This style muddies details so I mixed it with a simpler style trying to get it cleaner: CastStation Animation [LoRA: https://civitai.com/models/868249 Also used this to improve background quality: Forest Background [LoRA]: https://civitai.com/models/1311085 Style prompts: retro artstyle, flat color, anime_coloring, anime_source, anime_screencap, <lora:style_caststationanimation_illustriousXL-GLoRA:0.5>, <lora:RSTHTC_Style__Illustrious:0.7>, rsthtc2, <lora:ForestILL:0.4>, While inpainting I swapped the weights to favor the simpler style.

>>6024 Hopefully, it will help me learn game development and 2D animation. I'm testing some stuff, and I'm cautiously optimistic. I'm currently testing some sprite loras, flexibility, and checkpoints. There isn't consistency right now, but that's easy to fix later. I'm focusing on testing loras and creating a nice workflow.

https://civitai.com/models/1324976/il-high-cut-panties I got tired of Illustrious (and Pony. Like always I'll train this for it after I'm sure I don't need any adjustments) not knowing how to make "normal" or different panties (outside of a few that got tags like side-ties, thongs, and, for Illustrious only, briefs), so I decided to teach it how. What other styles of pantsu should I try making LoRAs for?

Gave OpenPose and the Openpose Editor extension a try. Also pretty simple to use. Just select OpenPose on ControlNet, choose an image of a full body character, select a preprocessor (dw_openpose_full), select a model, run the preprocessor and it creates a skeleton which can then be easily edited with the Openpose Editor extension. You can also save the skeleton for later use and skip the "image choosing" part. Pretty nice for some quick posing and you can even get a bit creative with extra limbs and multiple characters. I used a NoobAI-XL openpose model, but it works well with Better Days. NoobAI-XL ControlNet-OPENPOSE - https://civitai.com/models/962537?modelVersionId=1077649 Openpose Editor extension - https://github.com/fkunn1326/openpose-editor

This can help get a "cel-shaded" look with black shadows. Angel Cop Style - Illustrious [LoRA]: https://civitai.com/models/1277470 (artist:alkemanubis:0.7), <lora:angel_cop_IL:0.9>, ang3lc0pstyle, retro artstyle, anime_screencap, flat color, anime_coloring, anime_source, >>6134 >tagcomplete extension Installed this. Popup list is annoying at times when I don't want it, but I learned it doesn't only autocomplete booru tags. I can also use autocomplete to insert LoRAs into the prompt so I don't have to go looking for them.

>>5987 Bost em

>>6146 install the civitai helper extension too. it assigns metadata to your loras.

>>6024 see >>5931

>>5987 >>6147 There are prompts & thumbnails for 35 characters now. I put them in an archive. Character Prompts: https://files.catbox.moe/0yrgft.7z If you install the Stylez extension in A111, Forge or reForge, you can then extract the archive into the necessary directory, refresh the Stylez section of Stable Diffusion, then all the characters should appear with their prompt information. Tags are still imperfect. Plenty of tard-wrangling the AI. For example, Hat Kid's hat cannot be (purple top hat with yellow ribbon) as that puts yellow ribbons everywhere. Instead it must be (purple top hat with yellow band) to work & also needs (large headwear) with (small headwear) in negative to keep the hat BIG. Some characters need (chibi:0.7) or higher to generate as a loli properly. Some characters need more weight on their name tag. Still more experimenting to do. This is about as much as an Illustrious or NoobAI mix checkpoint seems to natively "know" about these characters while descriptive tags force it to stick closer to the correct appearance. LoRAs know far more, but you save disk space not relying on LoRAs for everything.

https://civitai.com/models/1324713/il-experimental-lucasarts-1995-96-cutscene-style Made a subtype of LucasArts style. Unfortunately, while it seems to works for men, it doesn't seem able to do women.

>>6151 Also going to be the last really "new" thing for at least a week. Got some Pony retrains queued, but everything left I want to play with requires more work and I can't keep up my streak of one a day.

>>5987 >If I edited the CSV file it uses in another application it stops recognizing the file & wipes everything so I couldn't even alphabetize it. That can't be, I've definitely edited that csv file manually a lot in notepad++. You must have missed a comma or a quotation mark or something.

good bye thread.