RevealLayer lands in ComfyUI: bounding-box guided occlusion-aware layer decomposition

TBG______ · reddit · 2026-11-07

The TBG team wrapped RevealLayer into a ComfyUI custom node that splits a flat image into multiple RGBA layers, with a detailed comparison of competing approaches.

Three routes compared

Qwen-Image-Layered: automatic, variable layer stacks, can decompose iteratively; but layer count/ordering/semantics are less explicit, and TBG's tests showed speed and quality issues
CLD: user-provided bounding boxes guide which regions become layers; great for graphic-design/poster compositions, but RevealLayer's authors report edge softness and residual artifacts in complex natural scenes
RevealLayer: bounding-box region guidance combined with an occlusion-aware diffusion process — it separates visible elements and estimates what's hidden behind them; this is the route TBG brought into ComfyUI

Why bounding boxes work well

Instead of hoping the model invents the right layer plan, you hand it a region to work with
Alpha channels give each layer transparency; the occlusion-aware part recovers covered content
Key caveat: hidden areas are model-generated estimates, not recovery of the original hidden pixels

Useful for canvas apps, compositing, and moving elements around. Node, workflows and links are on their Patreon.

Original post →

RevealLayer lands in ComfyUI: bounding-box guided occlusion-aware layer decomposition

More from Multimodal