🐯 YFG PixelArt

Pixel art is one of those effects that looks like it should be hard and is actually a party trick: take an image, shrink it way down, blow it back up with nearest-neighbor sampling, and the chunky blocky look appears by itself. That's exactly what this node does, and it's the rare ComfyUI effect node where you can understand 100% of the mechanism in one sentence.

The "YFG PixelArt" node (from the gonzalu/ComfyUI_YFG_Comical pack, adapted from XSS's original Mosaic code) takes an IMAGE in and returns a pixelated IMAGE out. Under the hood it's two torch.nn.functional.interpolate calls: a downscale to your chosen pixel grid, then an upscale back to the original resolution using nearest-neighbor. Done. The blockiness is just information loss made visible.

The two inputs that matter

• pixel_size (1–512, default 4) - how chunky the pixels get. This is the real "pixel size" of the result: 4 gives a subtle texture, 16+ is aggressive, and anything past 32 starts to look like a mosaic of a photograph. Push it and preview.
• interpolation_method - which resampling runs on the downscale: nearest, bilinear, bicubic, area, or nearest-exact. nearest (the default) gives the crispest, most authentic pixel-art look. area averages blocks and gives a smoother, antialiased feel. Everything except nearest blurs slightly, which some people prefer - but if you want true retro blocks, leave it on nearest.

The output is the same resolution as the input, so you can chain this straight into a SaveImage or an upscale without thinking about sizes.

When to use it

It's a fun, instant stylization - game-asset mockups, retro profile pictures, thumbnail art. It's also a genuinely useful analysis tool: pixelating an image before inpainting or compositing can hide generation artifacts in areas where you don't want them to be visible. But it's a one-trick node, and the trick is well-trodden - ComfyUI ships similar downscale-then-nearest workflows, and the "Nearest" ImageScale mode does most of this. The convenience is that it's one node with a single pixel-size knob instead of a two-node dance.

Installing

Part of the pack: ComfyUI Manager → search "YFG Comical", or

cd ComfyUI/custom_nodes
git clone https://github.com/gonzalu/ComfyUI_YFG_Comical

Restart ComfyUI. This node needs only torch + PIL, so it's the easiest piece of the pack to get running - no model downloads, no API keys. (The pack-wide cv2/matplotlib import issue can still take the whole pack down; if so, pip install opencv-python matplotlib.)

The honest take

For a quick stylized effect it's perfect - zero setup, one knob, instant result. Don't expect it to do anything cleverer than block-ify an image, because it doesn't need to. If you're after authentic pixel-art generation (the model kind, with palettes and dithering), that's a prompt problem, not a node problem - this is the cheap and cheerful version.