Color Auto Color Enhance (MD)

Ever generate something that's technically sharp but looks like it was shot through frosted glass? Muted colors, no pop, everything a bit gray even though the checkerboard is fine. That's the flat-color look, and mdColorAutoColorEnhance (one of the MdColorMod nodes) is a zero-brainpower fix for exactly that.

It's a straight port of GIMP's Colors → Auto → Color Enhance. GIMP calls that trick "chroma stretching": the image is converted to CIE LCh color space, and the chroma channel (how colorful each pixel is) is stretched so its minimum maps to zero and its maximum to full. Colors get pushed apart from each other - reds get redder, blues bluer - without touching lightness or hue. Unlike a saturation slider, which pushes everything uniformly, this stretches whatever range of colorfulness your image actually has. An image that's only slightly colorful gets a bigger relative boost than one that's already saturated.

What you actually touch

The node is deliberately tiny:

• image - your IMAGE tensor. That's it for required inputs.
• mask - optional. If you only want the effect on part of the frame (say, the subject but not the background), feed a mask here. Two things happen: the chroma range is computed from the masked region, and the result is blended per-pixel by mask strength.
• Output: image, same size, same channel count.

The "auto" in the name is doing real work: this is a per-image statistical operation, not a slider. No tuning to get wrong. When the mask is used, it also does per-image range calculation, which makes batch processing behave nicely - each frame gets its own stretch instead of one global curve.

Where it fits in a workflow

This is textbook post-processing, the deterministic layer the sampler hands off to. MdColorMod's pitch is that all of these nodes take masks and pass alpha through untouched, so you can drop this after the VAE decode and before any upscale. Pair it with a subtle film grain or a gamma bump and flat AI output starts reading as a photo - that whole de-AI playbook is covered in our post-processing notes, and this node is the "colors look flat" step.

Installing it

MdColorMod is a plain git install with zero Python dependencies - its requirements.txt is empty and everything is pure PyTorch. No models, no weights, nothing to download. ComfyUI Manager can find it by searching "MdColorMod", or:

cd ComfyUI/custom_nodes
git clone https://github.com/MdONeilsl/ComfyUI-MdColorMod

Then restart ComfyUI. It lives under the md/image/color category.

Gotchas worth knowing

The pack is new and small - as of writing it's barely a month old with a one-line README, so if Manager doesn't list it yet, the manual clone above always works. Two source-level notes: the mask should be (B, H, W) shaped for most nodes here, and if you feed a batch of images it auto-chunks on GPU to avoid OOM. Also it's GPL-3.0, which is fine for personal use but worth knowing before you ship a product built on it. If the effect looks too strong, there's no strength slider here - pull the mask out or use one of the manual color nodes in the same pack instead.