This is 8bitdiffuser! A model trained on ~2k images
This page is for the 64x model!
PLEASE when using this model, use one of the AnyLoRA forks, Qteamix, Angelbreed or Animekawa. it works best with stylized anime / cell shaded models, while it can work on many other models, it might not produce the "pixel perfect" results
When generating images scaling them down 8x will produce pixel perfect results! you can downscale them via billinear or nearest neighbor, but i have had the BEST results when using astropulse's Pixelize plugin for A1111, it is completely free and can run even on lower end devices.
make sure to use an 8X downscale factor in post processing!
the 64x model, while a bit worse at generating even pixels, still operates greatly when downsampling images to native size, it also works much better at full body character portraits. with all v3 versions and above, high resolutions (768x and 1024x) are highly reccomended
Note for versions v4 and above! i used some filtering on the dataset, so if it gives you problems with NSFW content, use the v3 or earlier models.
Note that even doe this page specifies a 64x model, u can generate different resolutions of pixel art by just scaling the picture!
256x256 will generate a 32x32 pixel art
512x512 will generate a 64x64 pixel art
1024x1024 will generate a 128x128 pixel art
etc etc...
best settings for this LoRA is:
Steps: 10-30 (On lower step counts, pixels will look very noisy, however if you downsample this to native pixel size this issue is resolved, increase step count if you want a great generation without downsampling)
CFG scale: 4-10
LoRA scale: 0.85-1.25 (the lower you will set the scale, the more it will look like traditional art and the pixels will become uneven, but the higher the scale is it will appear much noisy)
Sampling method: DPM++ 2M SDE karras
Recommended prompt: "pixel_art" (Add these to the start every generation, this isn't "required" but I found it gets the best results)
Resolution: 512-1024 (on lower resolutions, the pixels will be less even, on higher ones the artifacts will be less noticeable but will take much more time)
Description
Absolutely huge update. the training data has been completely overhauled,
the biggest change is that this can now work on higher resolutions! and much less style bleeding, i reccomend a size of 768x768!
FAQ
Comments (10)
Love this model, its great ! have you got any tips for training ? I'm struggling to get anything consistent when training on pixel art for some reason. Cheers
got a few, first make sure all the pixel sizes are consistant, so if in one image each pixel in the pixel art is 4 pixels wide, but in another its 3, it wont work. since the ai needs to pick up on the pattern for this i trained it all on 64x64 pixel art.
its also important the pixels arnt blur, the image needs to be sharp and upscale to 512x512 (or 1024x1024 if its an XL model) without blurring the image
@extraordicord_byrtexploit Cheers, helped a lot!
what are the permission for the model?
can you make game assets and sell the final product with this model?
Short answer: Yes! you are allowed to use the images this model generates in commercial products!
However if u do end up using the model, make sure to provide credit to this page on either your game credits or in the store page in which you will sell this!
Yeah. Just don't say anything. Easy peasy. Who gives a shit. Jesus.
I'm confused as to where in the process I should use this LoRa. Am I using it to generate a large image, moving that image to extras, and then using Pixelize to reduce it? I can't see anything in the extras page that indicates I could use a LoRa, there and I'm not sure what the LoRa is doing if I'm just using it on the large image. I do think the examples provided look very good, but please explain how and where I should be using this LoRa and what it does. Thank you!
Generate the image using the LoRA, the LoRA itself already generates pixelized images, but they can be a little off (some pixels a little bigger / smaller then others) to fix this you can then you can either pixelize with an extension or with any website (basically just scale down the resolution of your image by 8)
just a headsup, pixel art effect is achieved using "None", "Linear" or "Nearest Neighbor"; Using "Bilinear" will blur the result.
You can use ffmpeg to iterate over all images on a folder to resize them using the correct filter (Linear in ffmpeg) for training.
For me strangely billinear might be a bit blurred but it achieved the best results (except for k-centroid, I HIGHLY SUGGEST YOU USE K-CENTROID)
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.