>>2069802 >>2069939 I found the true culprit. [edit: fixed a few things]

Made a Z-Image Turbo LoRA. Not sure it works well but I don't really know how to use ZIT. Training data is released since it's all CC or PD. https://civitai.com/models/2261484?modelVersionId=2545599

>>2070593 ToT >>2069106

Made some glitchy style portraits of some of the best girlmons. The ones that don't require a lora anyway. Lilithmon, Beelstarmon, Angewomon, Ladydevimon and Lilimon. Happy New Year! Tags that may be of interest: common- grid, hexagon, abstract, glitch, vhs artifacts, cropped, cropped torso, outside border, portrait 1- horror, shades, squinting, cigarette 2- high contrast, anime screenshot, anime coloring, scanlines, dithering 3- scanlines, dithering 4- scanlines, horror, dithering, eye trail 5- scanlines, dithering, spoken heart >>2064942 It's not really a problem. He is just a one-eyed floating head, so naturally trying to get him to skateboard and do other weird things is going to mess him up. The profile thing is a bit odd, but not a big deal.

>>2019377 Someone else wound up making Gondola. https://civitai.com/models/2271387?modelVersionId=2556659

I couldn't sleep so I tried out this LoRA: https://civitai.com/models/1527049/secret-photo It struggles with taller aspect ratios so I had to improvise & crop it to 16:9 later. 0.5 weight - incase new style: https://civitai.com/models/714944 Forge couple prompts had artist tags: (artist:incase, artist:alex_ahad:0.6), >>2070884 Fug yeh :DDD >>2070777 checked Neat.

>>2070884 >>2071264 First merchandise, now this. smh 😔

>>2070884 >>2071264 Much needed. AI can generate neat backgrounds & Gondola is basically a small guy traversing various places (often with neat backgrounds). Tried it out. When you mix Gondola LoRA with style LoRAs you will struggle to keep correct form of the body. Some styles are worse than others for this as they're intended for styling humans. If you're blending stuff expect arms & shoulders to appear until you've balanced it enough to make it work. Lotta prompting to reduce unwanted body parts & characteristics. By default, Gondola was coming out too thick. I didn't want featureless blob feet, but didn't want detailed toes either so I tried to make a simple paw look. Character Prompt: <lora:gondola_IL:1.2>, Gndla, short, 1other, solo, no humans, brown fur, black eyes, no sclera, black nose, yellow muzzle, no ears, (no arms, armless:1.3), legs only, paws, (otter face, skinny, skinny legs:0.4), toon \(style\), sitting on log, (feet on ground:1.1), Negative Character Prompt: open mouth, tall, seamed legwear, large ass, thick thighs, pectorals, breasts, arms, neck, shoulders, collarbone, Style (1.0 weight): https://civitai.com/models/1340056/the-lord-of-the-rings-style-illustriousxl Background (0.7 weight): https://civitai.com/models/1311085/forest-backgrounds I only inpainted the background for greater details. Gondola's design is so simple it doesn't really demand greater detail via inpainting, which didn't improve it much regardless. Maybe with simpler styles inpainting Gondola would make lines cleaner, but with complex styles not worth it IMO. This was the only gen where he is leaning forward. With the mix of LoRAs I used if I tried to make him lean forward in prompt it would create arms to rest on his legs so I just had to get a forward leaning pose by random chance.

How well does AI handle small-tiny unit/character sprites. I wanted to make some modded units for a game but I am feeling too lazy to make my own sprites this time around and was mildly hoping AI can at least manage decent-ish placeholders.

Civitai down so I cannot find & link the LoRAs I used here. >>2072524 This has been mentioned in past threads. Pixlization filter extension: >>>/ais/6121 Pixel art LoRA examples: >>1531282 >>1633105 I don't mess with pixelization much, but maybe that anon who does will tell you more about it when they're around.

I'm using grok to inpaint some images and its muich better than i expected, at very least much better than novelAI, only downside it's censored.

>>2072467 My (limited) testing with creatures says "pokemon (creature)" is stronger than specifying the weird features, but it needs to be baked into the trained LoRA and this one doesn't use it. Could probably still try making a better one myself, but would need more training data (I need 30+ and have 10)

it still bothers me because i want to know what the fuck he saw in it, but given the allegations about stephen hawking liking to watch nude midgets write equations of a blackboard that's to high for them to write comfortably(on epstein's island), could i request something of the sort but with glasses-wearing gobbos/midna?

>>2072868 Weird story, bro. >i want to know what the fuck he saw in it <proceeds to request images depicting the fetish lol I think "stretching" "tiptoe" & "arm up" tags could get the pose, but I'm unsure about getting them writing specifically on a chalkboard mounted too high on the wall.

>>2068566 >>2072467 Beautiful. >>2070777 Since you don't have Sakuyamon and Fairimon... https://civitai.com/models/1449710/penetration-gesture-concept >>2072524 Not well. I don't think it's usable without a lot of work and at that point might as well just DIY from scratch. Online models like Gemini and Grok, if they can do proper pixelart, don't give you raw high-resolution images so it's pointless. >>2072703 >>2072524 None of the extensions work with transparency afaik which is also problem you have to work around. Things that might be of use: https://github.com/mrreplicart/sd-webui-pixelart https://github.com/AUTOMATIC1111/stable-diffusion-webui-pixelization https://civitai.com/models/1631459 https://civitai.com/models/1492329 https://civitai.com/models/1686124 https://civitai.com/articles/19902 If you already know pixelart, then simply generate the images, downsize to a factor 8 and use as reference in Aseprite for actual pixel perfect art. That'd be the workflow imo.

>>2072524 Making sprites is not really something AI is capable of yet unfortunately It can do pixel art looking faces and artwork and such but that's about it

https://civitai.com/models/2206328/zit-golden-sun-official-art-style-610ths-trained?modelVersionId=2553625 Tried training full version of Golden Sun style ZIT LORA now that ZIT training isn't unstable, but seems way worse than the one that only got 6/10ths of the way trained before erroring out (that's the third and forth image). Should have cute, obscure, girls again soon (already got four trained and have dataset for a fifth almost ready)

>>2072868 >>1875637 Had to generate pose with no blackboard, then add blackboard manually to get the height. Wanted to fill the corner space with some onomatopoeia & moon runes just look better so しくしく is Shiku Shiku meaning gentle sobbing. Kemokino style [0.9 weight]: https://civitai.com/models/1537195 Dunno what Midna LoRA I used, maybe this one [1.1 weight]: https://civitai.com/models/1288810 Can't expect it to get Midna's color pattern exactly right. It practically never does with style LoRAs.

1girl is life. >>2073378 lul

>>2072766 Was thinking if you had a 3D model of Gondola you could capture any angle & if it were rigged you could even get poses, but the only model I could find is for 3D printing with no texture. https://www.thingiverse.com/thing:4281704/files The last time I modeled/rigged a Gondola was a few years ago in 3DS Max (which I don't even have installed anymore), but the model didn't have feet, it was too wide & flat in shape. I found an old animation test of it, but who knows where the files went. That Gondala I made only looks right if I horizontally squash the video. IIRC I made the model too wide to accommodate a pre-made skeleton that could do path walking animations. I guess the video could be usable if squashed horizontally & cropping out the no-feet stubs. I mean, he's turning & moving at least & there's over 100 frames there.

https://civitai.com/models/2264161/il-allison-mccourt-power-pack?modelVersionId=2548599 Allison McCunt McCourt from original 1980s run of Power Pack . No-Prize for guessing what two other LoRAs I've got in the pipeline (though I have another unrelated character as well ready to go too). >>2074550 Characters always posed the same way (most common with 3D models) have a nasty habit of baking in that posing (indeed, it's the main difficulty in making stuff from a 3D model). Though I suppose Gondola doesn't really have any over poses..

>>2071758 Another Peach gen. 0.9 weight - incase new style: https://civitai.com/models/714944 0.7 weight - girthy benis: https://civitai.com/models/1371020

Made a few portraits of the gravity shifters and angels from Gravity Rush. Kat, Raven, Durga and Kali. Tags that may be of interest: common- sharpie art, hatching, dithering, halftone, portrait 1- outside border, thumbs up 2- bored, crossed arms, ellipsis 3- :d, thumbs up 4- confused, question mark 5- bags under eyes, squinting, head tilt Raven: https://civitai.com/models/1398570/commission-illustrious-gravity-rush-raven Kali: https://civitai.com/models/1623259/kali-angel-gravity-rush-2-illustrousxl Durga: https://civitai.com/models/1623267/durga-angel-gravity-rush-2-illustriousxl >>2072467 That's a very nice background. >>2072933 Pretty cool. I keep forgetting about Sakuyamon for whatever reason.

>>2075635 🍄💦🍑 Image-2-Video + Vocal gen edited together with sound effects. Voice-cloned Peach from Mario Party 10. Mario uses existing clips as Zonos-TTS couldn't do a Mario impression. Creampie [WAN-Video LoRA] - 0.6 weight: https://civitai.com/models/1941545 SmoothMixWAN replaces the incase style face with whatever anime style it knows so I can't keep every detail of the original image. Only did 2 sections since it was losing too much detail on the 3rd section.

https://civitai.com/models/2266843 Julie Power, from the original 1980s run of Power Pack (I've got made a model for the version drawn by Gurihiru for the 2000s miniseries series). Picked up a very neat comic book style. Maybe I should make a style LoRA for this...

https://civitai.com/models/2266311?modelVersionId=2550955 Had to finish some stuff before I can train Katie (it turns out that when a character's power makes her glow and she's carried around by her older sister, it's a pain in the ass to get good data), so here's Aspara (fused with Water Shaman Seso) from Breath of Fire II . Actually had data for this since April 2025 and can't remember why I never used it (best I can think of is either wanting more or not wanting to clean it). Came out well for what I worked with, and it really shows what happens when a LoRA is trained on such disparate appearances it has little to no inherit style (especially when contrasted with Allison and Julie, which have a very distinct style inherently by being trained exclusively on one source)

>>2069034 >Third Image. Is it more of an oppai loli or a shortstack when you prompted it? I'm thinking of uploading to atfbooru if it's the former.

I'm really liking the "sideless outfit" tag. I daresay it's one of the most cultured creations of mankind.

https://civitai.com/models/2284953/il-katherine-katie-margaret-power-energizer-power-pack-earth-616?modelVersionId=2571608 And to no-one's surprise after Allison and Julie, here's Katie Power. Note: Civitai is currently fucked and can't rate new images, and without rated images a model can't be seen. I think the DL page part works if you have an account and a direct link, but not sure. Should work when they unfuck their shit.

>>2081466 I was thinking fat titty short stack.

>>2082945 For me It's the star dress Rosalina. Nice.

Created smut photo of Raven using Illustrious CyberRealism which is so-so at prompt adherence. Some tags needed a ton of weight to work & some LoRAs break outputs. Expected inpainting to butcher things, but it did fine. Still photoshopped a bit to get details I wanted. After that, played with the photo a lot using SmoothMixWAN-Video while trying different LoRAs. Took ages to get decent video gens, vocal gens, gather sfx, then ate more time editing it all together. Did a lot of vocal gens. Featured voice clones (spoilered if you wanna guess the crowd voices): link (CDI), Preecha (JP) (fatal fury), Amalia (FR) (wakfu), Carl (ATHF), Brimstone (valorant), Shulk (JP) (xenoblade), Marie Kanker (EEnE), Coco Bandicoot (CTR) & Raven (teen titans) >>2082945 Cultured as fuck.

1 & 2 - https://civitai.com/models/1514668 3 - https://civitai.com/models/56829 4 - https://civitai.com/models/2244268 5 - https://civitai.com/models/2261252

>>2086429 >for me it’s Disgusting cuckchan newspeak.