this is checkpoint that i mix it from various models and LORAs to create best pixel art image.
best resolution at 512x512, 768x768, 1024x1024, 512x768, 768x512 but feel free to experiment.
full model version available on my huggingface: https://huggingface.co/megaaziib/aziibpixelmix
order for commission: https://www.fiverr.com/s/kLNeb8y
use pixelart extension to downscale to 8 and upscaling again for more authentic pixel look.
https://github.com/mrreplicart/sd-webui-pixelart.git
download my workflow for comfy ui user:
FAST VERSION: https://openart.ai/workflows/XkwkHIWGhMLWxQuBIsd1
https://openart.ai/workflows/-/-/09EGyt3ZOBM9kD4ZZGP5
if you like my stuff and use it on commercial project, please consider supporting me on Kofi
for Fast Mode version, the cfg scale need to be 1 or 2 and minimum step is 4 or 8 or else the image will look worse.
Description
WARNING!!!! you must follow this setting or the image will look worse
CFG Scale - 1 or 2
Sampling Steps - 4 or 8
this model version is using LCM (Latent Consistency Model) so the result will be faster than regular stable diffusion.
FAQ
Comments (19)
Are you willing to upgrade this model? Or are you willing to mix this model in different versions with different picture styles? I'm asking because I really like it!🥰🥰🥰
sure, i will upgrade with other checkpoint or with other loras to improve it., right now i mixed it with LCM lora (latent consistency model) on Fast Mode, so it can generate more faster pixel art only 4 or 8 steps and cfg scale only 1 or 2. you can try on Fast Mode version.
@aziib Thank you so much for making this model. It's a model that I've been really looking for (I think I've been struggling for about a month.) It's the most perfect pixel model I've ever seen that the pictures are exactly square. The regret is that the painting is not colorful, and I am really excited to hear that you will compensate for that. I hope it will be a more colorful version up by mixing various models and Lora. I sponsored Ko-fi for you.🥰
(As an aside, I promoted your model to a community I know :) I would appreciate it if you could continue to create better models.)
@jss31776286 thanks. i will try to train lora with this model and if it's work, i will combined with the lora that i trained., and mix with other lora for new version.
@aziib Pixel art still uses this model well even now that nai v3 came out. This is the best model I've ever seen in pixel art! Just to be sure, do you have any plans to transplant this model to SDXL version as well?
Simply amazing! If the user of this model had shrunk the image to 1/8th size with "nearest neighbor" and then scaled it to the appropriate size using the same algorithm, the results would be even more accurate.
This model brings a special non-existent memory to middle-aged men around 40 years old. For some reason, the sexy NES/SNES-style images bring us nostalgia, something that should not have existed when we were teenagers. If you are my age, this is definitely a model you should try.
There's a WebUI plugin to do that nearest neighbor scaling automatically.
https://github.com/mrreplicart/sd-webui-pixelart.git
@Jellai yes i'm using that too, that is really useful.
thanks.
Excellent Model
Awesome model. I could see the model itself being incorporated into a procedurally generated game and the game would still feel somewhat consistent
This is the best pixel art model I've ever seen!
You really done it, sir. I've been experimenting with quite a few LORAs and extensions for pixelization. Got some good material for CGs mostly, but this, for sprites, is just absurd. Keep on the good work. Great asset to the community
Excellent performance in different angles and poses. T2P within TILE will have very good result.
What a miracle this model is! Without requiring a trigger word, it creates a beautiful cool retro pixel art picture from any prompt. It has great outstanding variability and can create as if anything in this style!
I love it!
All of my outputs are coming out as 64x64 pixel art, how can I upscale to have the output make higher res art?
Not only produces this model an almost pixel perfect 16-bit style it also works with almost any lora i could throw at it! Awesome Work!
Can this work with any controlnets? I can't get any to work...
I get redirected if I try to open openart.ai links. Could you share the ComfyUI workflow on the different place?