Yep. It will create something that at a glance looks like pixel art, then you zoom in and find anti-aliased rectangular pixels made out of three different shades of the main color, a weird line, and no clear grid.
A whole song and dance to post-process a broken pixel art image into a fixed grid. I’d rather just have talent or pay a human being that will make art in the first place, thanks.
Also, the example shown would be a terrible image to use as an asset, by the way. It’s a clear example of not pixel art, rather, 2D art forced to be interpolated in a “pixel like” fashion (aka what happens if you do a few nearest neighbor passes and then a fixed grid downsampling)
The person above you is wrong, while a generalist model will struggle to make valid pixel art, either a specially trained model or one with elements to force a specific structure could make generative pixel art.
AI is quite bad at pixel art in fact.
It understands things must look roughly made by squares, but it doesn’t understand what pixels are and their fixed grid.
To be fair, humans are also bad at it these days, judging by the amount of mixels.
They can base their pixel art off ai art, that resembles pixel art. Would still count as “no ai in the game” I guess.
even if it is exclusively trained on pixel art?
Yep. It will create something that at a glance looks like pixel art, then you zoom in and find anti-aliased rectangular pixels made out of three different shades of the main color, a weird line, and no clear grid.
If you use a pixel art export node in ComfyUI that won’t be a problem. There’s a whole guide about it here:
https://inzaniak.github.io/blog/articles/the-pixel-art-comfyui-workflow-guide.html
A whole song and dance to post-process a broken pixel art image into a fixed grid. I’d rather just have talent or pay a human being that will make art in the first place, thanks.
Also, the example shown would be a terrible image to use as an asset, by the way. It’s a clear example of not pixel art, rather, 2D art forced to be interpolated in a “pixel like” fashion (aka what happens if you do a few nearest neighbor passes and then a fixed grid downsampling)
The person above you is wrong, while a generalist model will struggle to make valid pixel art, either a specially trained model or one with elements to force a specific structure could make generative pixel art.
This seems like a valid example - https://pixel-art.ai/