Color Adjust

IsekaiColorAdjust is the finishing-touch node: brightness, contrast, saturation, and sharpness, all on one card, all with a slider each. If you've ever decoded an image, stared at it, and thought "that's 10% too washed out," this is the node for that moment. It's the last tweak before save - not the thing you build a workflow around.

How it works

It's a thin wrapper over PIL's ImageEnhance class, which does exactly what the four sliders claim. Each of the four controls is a multiplier centered on 1.0: 1.0 means "leave it alone," 0.5 means "half as much," 2.0 means "twice as much." The range is 0–2 on every control, and the defaults are all 1.0, so the node is a no-op until you touch something. That's the right design for a tweak node - you can drop it into an existing graph and nothing changes until you want it to.

Brightness scales overall luminance, contrast pushes dark and light pixels apart from the mid-point, saturation boosts or drains color intensity (0 = full grayscale), and sharpness applies Pillow's edge-sharpening pass (below 1 it actually softens). If you want the "photographed, not generated" look, the classic move is a touch of saturation reduction plus a nudge of sharpness - the community's realism threads keep rediscovering that combination.

The inputs

• image - the image to adjust.
• brightness, contrast, saturation, sharpness - each 0–2, default 1.

Output is a single image tensor. Wire it between your VAE decode / post-processing and a Save node. If you're stacking it with IsekaiLevels or IsekaiColorFilter, do levels first (fix the tone) then color adjust (fine-tune the mood), since levels re-stretches the histogram and would undo some of what saturation does.

Installing it

Same pack, same story - it ships in isekai-comfy-node:

ComfyUI Manager: search "isekai" → install isekai-comfy-node → restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/isekai-sh/isekai-comfy-node
cd isekai-comfy-node
pip install -r requirements.txt

Restart and look under Isekai → Image/Blend. Dependencies are Pillow and requests only; torch and numpy are already in your ComfyUI environment. No models, no downloads.

Common issues

Honestly, there's not much to trip on. It's a PIL passthrough - the only real failure mode is the pack-wide batch caveat: a batched image only gets its first frame adjusted, because the tensor-to-PIL helper takes frame zero. Single images are completely safe.

The one thing worth understanding is that these adjustments run at 8-bit precision after the tensor converts to a uint8 PIL image. If you push contrast or brightness hard (1.6+), you'll see banding in gradients that a float-precision node wouldn't produce. Keep the adjustments moderate and you'll never notice; crank them and it's the same lesson every image-processing tool teaches - extreme gain exposes the bit depth.