TBG RevealLayer brings occlusion-aware layer splitting to ComfyUI: one flat image becomes separate RGBA layers
TBG RevealLayer is a free ComfyUI custom node, released on 7 November 2026, that runs the RevealLayer research model to turn one flat image into separate RGBA layers. Objects you box, by hand or with SAM3 text grounding, become their own layers, and the model fills in what was hidden behind them.
What it does
It outputs a foreground layer, a reconstructed background, one RGBA layer and alpha mask per selected object, and a composite, with up to ten object boxes per image. Two workflows ship with it: manual boxes, or SAM3 grounding where text such as `person:2, handbag:1` proposes the boxes for you to edit.
Worth knowing
It runs on FLUX.1-dev (FP16, BF16 or FP32) with its CLIP-L/T5 encoders and VAE; FLUX.2 and quantized weights are not supported, and FLUX.1-dev has its own licence terms. The first run downloads about 697 MB of RevealLayer files (Apache 2.0, from 360 AI Research); the model works at a 1024-pixel long side, and hidden areas are generated, not recovered pixels. The node code is MIT.