MLiang White Balance (Legacy)
The legacy white balance node — an older name for the same cast fix
- image
- image
"MLiang White Balance (Legacy)" is what happens when an author renames their own node and decides not to orphan everyone who already saved a workflow with the old name in it. This node is the white balance fix from the same pack, registered under the pack's pre-rename "Codex" class name. It does exactly what the modern MLiangWhiteBalance does, because it is that node under the hood.
What it is
Look at the source and the entire legacy implementation is a subclass plus a flag:
class CodexWhiteBalance(MLiangWhiteBalance):
DEPRECATED = True
Same five inputs - red, green, blue channel multipliers (1.0 neutral, 0–3 range), normalize_luma (on by default, keeps overall brightness steady while you correct the cast), and strength (blends the correction back toward the original). Same single IMAGE output. Same math: multiply the channels, re-normalize luminance if asked, clamp to 0–1, keep any alpha channel. DEPRECATED = True is just a signpost; the node still registers and runs identically.
You'll see it for exactly one of two reasons: a workflow you loaded references CodexWhiteBalance by name, or you're poking through the node menu and spot both the modern and legacy entries. Both are expected - the pack intentionally ships all four classes so old and new workflows resolve.
Modern or legacy?
If you're building fresh, drop in MLiangWhiteBalance and skip this one - no functional difference, but the modern name matches the README and the example workflow. If an existing graph already uses the legacy node, there's nothing to fix. Same inputs, same output, same behavior; re-wiring the graph to swap names is pure cosmetics with a small chance of breaking something.
For the actual how-to - fixing a green cast by dropping green toward 0.9, why normalize_luma is the clever part, and why you should switch it off if you want the channel scaling to intentionally darken - the modern node's page covers all of it. Installation is the same pack either way:
cd ComfyUI/custom_nodes
git clone https://github.com/dingmuliang/comfyui_mliang_color_correction.git
Restart ComfyUI after cloning. No models to download, no extra Python packages; the whole pack is one file of torch.
A final word of caution that applies to the whole pack, legacy included: this is a tiny one-person project - a single commit, a handful of stars, and a WeChat Pay tip jar in the README. Nobody in the community is really talking about it, and these pages have zero search traffic. It's clean code that does one job well, so that's not a knock. Just don't build a mission-critical production pipeline on a dependency this young without backups.
Inputs (6)
| Name | Type | Default | Description |
|---|---|---|---|
| image | IMAGE | — | |
| red | FLOAT | 1.000–3 | — |
| green | FLOAT | 1.000–3 | — |
| blue | FLOAT | 1.000–3 | — |
| normalize_luma | BOOLEAN | true | — |
| strength | FLOAT | 1.000–1 | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| image | IMAGE | — |