Siddharth Kumar Sah 93dd37017c feat: add Ultra quality mode with BiRefNet-matting for people photos ...

Adds a new "Ultra" quality tier for People subject type that uses
BiRefNet-matting (ONNX, 928MB) for true alpha matting instead of
binary segmentation. Produces per-pixel transparency for hair wisps
and fine edges that standard models miss.

- Custom rembg session class loads BiRefNet-matting ONNX from GitHub releases
- Zero new Python dependencies (reuses existing onnxruntime)
- Model pre-downloaded in Docker build alongside existing models
- Ultra option only visible when subject is People
- Falls back to Best when switching to Products/General

2026-04-12 18:09:59 +08:00

..

ai

feat: add Ultra quality mode with BiRefNet-matting for people photos

2026-04-12 18:09:59 +08:00

image-engine

feat: overhaul remove-background with effects pipeline, consolidate color tools

2026-04-12 17:53:16 +08:00

shared

feat: overhaul remove-background with effects pipeline, consolidate color tools

2026-04-12 17:53:16 +08:00