Paint-Anything: Unified Any-Color Control for Image Generation and Editing

The authors propose Paint-Anything, a system that learns a shared hex-prompt interface for image generation and editing through object-level color supervision. They construct a data pipeline called Paint-500K combining real image data with pure-color anchors at high-noise timesteps to mitigate shadow limitations. They also introduce ACBench, comprising ACBench-T2I and ACBench-Edit, to evaluate performance. On FLUX.2-4B, Paint-Anything improves ACBench-T2I and ACBench-Edit scores by 85.3% and 28.3%, respectively, relative to the base model.

Paint-Anything learns a shared hex-prompt interface through object-level color supervision.

The Paint-500K dataset is constructed from real images complemented by pure-color anchors at high-noise timesteps.

On FLUX.2-4B, Paint-Anything improves ACBench-T2I and ACBench-Edit scores by 85.3% and 28.3%, respectively, relative to the base model.