load color config

Semantic-segmentation and DensePose outputs aren't masks by themselves - they're color-coded images where each region's color stands in for a category (a body part, a piece of a scene). load_color_config loads the lookup table that translates those colors into meaning, which is the first step before you can actually turn one of those color maps into a usable mask elsewhere in this pack.

It's part of ComfyUI-WJNodes' color toolset, alongside filter_DensePose_color (which consumes this node's output directly) and a couple of color-segmentation nodes documented in the same README, all inside a large personal utility pack from GitHub user 807502278.

How it works

Pick a config_file - the options cover ADE20K (a widely-used semantic-segmentation color scheme) at two format versions, plus DensePose variants rendered with different color palettes depending on which upstream tool produced them: plain DensePose, a CivitAI-flavored version, and a MagicAnimate-flavored version, in both Parula and Viridis colormaps. Different DensePose implementations render their UV maps with genuinely different palettes, so this isn't a cosmetic choice - picking the wrong config file means the color-to-region lookup won't match your actual input image at all.

The inputs and outputs that matter

• config_file (default ADE20K_v1.1.json) - which color scheme to load. Match this to whatever actually produced your color-coded input image.
• version (default 1.0) - a separate version selector from the config file's own version number in its filename; worth double-checking these agree with what your config file actually is, since the node's own default combination (ADE20K_v1.1.json file with version set to 1.0) is a mismatch you might want to fix explicitly rather than leave on default.
• tag_language (default EN, English or Chinese) - controls the language of the tag_list output text. The node's own description notes this setting only has an effect on config version 1.1 and above.

Outputs: Color_list and Color_dict (the color lookup in two shapes - Color_dict is what filter_DensePose_color expects), and tag_list (the human-readable labels as text).

How to install it

Via ComfyUI Manager, search "ComfyUI-WJNodes." Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/807502278/ComfyUI-WJNodes.git

Restart ComfyUI. No model downloads needed - this node loads a bundled JSON config, it doesn't run any detection itself. You still need a real DensePose or ADE20K-style segmentation preprocessor elsewhere in your workflow to actually produce the color-coded image this config is meant to interpret.

Common issues & troubleshooting

Filtered or segmented output looks like nonsense. By far the most likely cause is a config/source mismatch - DensePose images from different tools use different palettes (Parula vs. Viridis, and further variation between the plain, CivitAI, and MagicAnimate variants). Confirm exactly which preprocessor generated your input before picking a config file, rather than picking one and assuming it'll line up.

tag_language doesn't seem to change anything. Check the config file's own version - the node's description says this setting is only honored on version 1.1 configs and above; it's a no-op on 1.0-era files.

version and config_file disagree. The node's own default state loads a v1.1 config file while version sits at 1.0 - that's not necessarily broken, but it's worth setting both deliberately rather than trusting the out-of-the-box combination if version-specific behavior (like tag_language) matters to you.

No community reference for these specific colormaps. ADE20K and DensePose's implementation-specific palettes aren't something this pack's docs (or ours) cover in depth - if a mapping still looks wrong after matching the config to your source, the fastest check is visually comparing a few known regions in your color image against the tag_list labels.