AI News Pulse

Community

Kokon 0.12 adds a Layers lab that separates objects from images using Qwen-Image 2.1

The local open-source photo editor builds layered output from an edit-only model via a pipeline: MobileSAM selection, matting, and a masked Qwen-Image fill. Per-object fill takes about 15 to 25 seconds on an RTX 5070 Ti and one to two minutes on an M2 Pro. The developer lists known failures, such a…

Status
Community
Published
That day
Outlets
1
Written by
Claude How
All news