APNext Rough FX

The pack bills this one as adding "texture and roughness," and the way it gets there is by cutting your image down to a limited color palette. Where a normal render uses millions of subtly different colors, APNext Rough FX quantizes it to a few dozen, which gives you that flatter, chunkier, poster-or-screen-print quality - the banded, reduced look of a risograph print, a retro game, or a cheap paper poster. If you want a stylized, less-photographic finish, this is the knob.

Image in, image out, after VAE decode. No model, no API key.

How it works

The core control is color_count: it reduces the image to that many distinct colors (from as few as 4 up to 64). Fewer colors means flatter, more graphic, more obviously stylized; more colors keeps it closer to the original with just a hint of banding. The aa_factor supersamples during processing to control how clean or crunchy the edges of those color regions come out - higher smooths the boundaries between color bands, lower keeps them hard. resize_method picks the resampling filter used in that process: BICUBIC and LANCZOS are smooth, NEAREST is hard-edged and blocky (reach for it if you want a pixel-art crunch rather than a smooth poster).

Between those three controls you can slide from "subtle print texture" to "8-color screen print," which is a look you'd otherwise fake in an image editor.

Inputs and outputs that matter

• images - your decoded image.
• color_count (4–64, default 24) - the main dial. Low for a bold graphic-poster look, high for a light touch. This is where the effect lives.
• aa_factor (1–8, default 4) - edge smoothness between color regions. Higher is cleaner.
• resize_method - LANCZOS/BICUBIC for smooth, NEAREST for a hard, blocky, pixelated finish.

The single output is image.

Installing it

ComfyUI Manager: search comfyui_dagthomas, install, restart. Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/dagthomas/comfyui_dagthomas
cd comfyui_dagthomas
pip install -r requirements.txt

Runs on the pack's image dependencies. No API key, no model files.

Common issues

The obvious one: set color_count too low on a photographic image and detailed areas - faces especially - fall apart into ugly color blobs. If a portrait looks diseased, that's too aggressive a palette; raise the count. This effect flatters graphic, poster-style, and illustration subjects far more than it does realistic photos, so match it to the content.

If you're chasing a smooth poster and getting jagged edges, you've probably got resize_method on NEAREST - switch to LANCZOS and raise aa_factor. And because this is a heavy stylization that rewrites the whole palette, run it near the end of your chain and be careful stacking it with color grading; grading a heavily-quantized image tends to just shift the few remaining colors around rather than doing anything subtle. It's a finishing look, not a base pass.