BD Channel Merge

Game-engine texture packing is a game of Tetris with channels: albedo here, roughness there, a mask stuffed into the alpha. The pack's own "Game-Engine Packing" workflow isolates parts, balances luma, crops them to a shared pivot, then packs them into R/G/B channels of one texture. BD ChannelMerge is the tool that does that packing. It blends a source image into one or more specific RGBA channels of a base image, with a proper set of Photoshop-style blend modes - so you can paste a character into the red channel, or multiply a roughness map into the green channel, one channel at a time.

The standout feature is use_source_alpha (on by default): the source image's own alpha acts as a soft blend mask. Transparent source pixels leave the base channel untouched, opaque pixels apply the blend at full strength. Which means you can paste a character cut out on a transparent background into a single channel without its background pixels contaminating the channel. That's the difference between "packing a texture" and "painting a rectangular smudge into the R channel." The node description's example is exactly this: composite a character with alpha onto one channel, background be damned.

The inputs that matter

• base - the image being modified. RGB or RGBA; channels you don't target pass through unchanged.
• source - the image to merge from, auto-resized to match base.
• target_channels - where to write: single (R, G, B, A) or multi (RG, RB, GB, RGB, RGBA). Multi maps source channels 1:1; single collapses the source to a scalar first.
• source_from - how to collapse the source for single-channel targets: luma_bt709 (default, matches the skin shader), a specific channel_R/G/B/A, average, or max_rgb. Ignored for multi-channel targets.
• blend_mode - replace, add, multiply, screen, overlay, darken, lighten, subtract, difference, soft_light. Default is replace; multiply is your friend for shading one channel by another.
• strength - overall opacity, 0–1. Combined multiplicatively with the source alpha and any external mask.
• mask (optional) - an additional blend mask layered on top of everything.

Outputs: image - the modified base, same resolution, RGBA if the base was RGBA or if A is a target channel.

How you'll chain it

Because each pass modifies one channel and hands the result back, you can build a texture pack progressively: base → merge character into R → merge another into G → merge a mask into A → out. It's the pack's "Channel Operations" workflow in node form, and the node description explicitly encourages chaining several in a row to build up a texture one channel at a time. It's a deterministic pixel op, not a model - instant, repeatable, and exactly the right tool instead of burning a diffusion pass on work a channel map does perfectly.

Installing it

Part of BizaNator/ComfyUI-BrainDead (BrainDeadGuild). ComfyUI Manager → search "BrainDead" → install, or:

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

Restart; it's under 🧠BrainDead/Segmentation.

Gotchas

Three things to keep straight. First, if your source has no alpha channel, use_source_alpha does nothing - you get a hard-edged blend, so check your source has real transparency when the "no background contamination" behavior matters. Second, for single-channel targets, the source is collapsed before blending - pick source_from deliberately, because "what does red mean here" is exactly the decision that changes the result. Third, remember the output inherits channel count from the base and targets - target A and you're now RGBA even if you started RGB, which downstream nodes may or may not expect. And if a blend looks like it did nothing, check strength and whether invert_alpha_mask flipped your mask's polarity on you.