1hew-ComfyUI-1hewNodes/web/docs/1hew_DetectRemoveSolidBGUnMult/en.md

Detect Remove SolidBG UnMult - Solid-background UnMult Keying

Node Purpose: Detect Remove SolidBG UnMult targets footage shot on a flat background: it estimates the background color from the image border, unmultiplies channel-wise to derive alpha, then optionally suppresses low alpha with threshold and composites the result onto background_color. It outputs both the foreground image and an alpha mask. Auto background estimation and dithering are fixed (auto background on, dither off), so no extra switches are exposed.

Inputs

Name	Port	Type	Default	Range	Description
image	required	IMAGE	-	-	Input image batch; 1 / 2 / 3 / 4 channels are accepted.
threshold	-	INT	0	0-255	8-bit alpha black point. 0 disables cleanup; higher values suppress low alpha and remap the remaining alpha into [0,1]; 255 makes the output fully transparent.
background_color	-	STRING	empty	color name / HEX / RGB / sm / empty	Post-processing background. Empty keeps transparent RGBA; sm smart-selects black / white / #808080 per image; any other value uses the Image Alpha Split color syntax.

Outputs

Name	Type	Description
image	IMAGE	The unmatted foreground: RGBA when background_color is empty, RGB composite otherwise.
mask	MASK	Float alpha mask in the 0~1 range (input alpha applied, threshold honored).

Features

• Background estimation: samples the border strips, votes on quantized joint-RGB clusters, and picks the most spatially supported color per image.
• UnMult: chooses one-sided or two-sided channel unmultiply based on whether the estimated background leans light or dark, then applies a gamut-safe unpremultiply that clips negative channels and jointly scales positive overflow, so values stay in range without one negative channel reverting the whole pixel to the observed RGB.
• White-background gain: when the estimated background is nearly pure white, alpha receives an adaptive gain (median-driven, capped at 2.5) that improves translucent edges.
• threshold black point: remaps matte alpha with (a - t) / (1 - t) and zeroes pixels below the threshold while preserving any input alpha.
• Post compositing: background_color only affects the final image, never the background estimate; sm reuses Image Alpha Split's smart background choice.

Typical Usage

• Studio / flat-background keying: connect image and raise threshold from 0 to remove edge haze.
• Keep transparency for downstream compositing: leave background_color empty and use the RGBA image plus mask.
• Get a ready-to-use background immediately: set background_color to sm, white, and so on.

Notes & Tips

• No model inference and no weight downloads; the node is pure tensor math.
• The background model assumes a single flat color, so gradients or multi-color backdrops may leave edge and translucency errors; threshold or background_color can mitigate them.
• Inputs that already carry alpha are supported as well: the input alpha takes part in the final mask.
• The background_color / sm syntax matches Image Alpha Split exactly, so the two nodes are interchangeable.