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

• Field: Computer Vision
• Authors: Ji Xie, Dewei Zhou, Xinyu Huang, Zhennan Chen, Xun Wang
• arXiv: 2609.20816

Key points

• Goal: Professional design requires "any-color control" — the ability to specify an object's target color with any 24-bit hex value for both image generation and editing. Prior work on color generation, editing, and colorization typically relies on dedicated color representations or specialized inference procedures.
• Approach: Leveraging the fact that even compact LLMs can associate hex values with color semantics, Paint-Anything learns a shared hex-prompt interface for generation and editing through object-level color supervision.
• Data: A data pipeline builds Paint-500K from real images via object grounding, perceptual color labeling, and editing-pair synthesis.
• Training trick: Since shadows make real-image labels only approximate colors, supervision is supplemented with solid-color anchors whose pixels exactly match paired hex values. These anchors are used only at high-noise timesteps; low-noise training is left to natural images.
• Benchmark: The authors introduce ACBench (ACBench-T2I and ACBench-Edit) to measure object-level hex color fidelity in both tasks.
• Results: On FLUX.2-4B, Paint-Anything improves ACBench-T2I and ACBench-Edit scores by 85.3% and 28.3% respectively; ablations support the training scheme, and it achieves the highest average CompColor score among compared methods.

Abstract (original)

Professional design requires any-color control: the ability to specify an object's target color with any 24-bit hex value for image generation and editing. Prior work has explored color generation, editing, and colorization, but often relies on dedicated color representations or specialized inference procedures. Advances in large language models offer a simpler starting point: even compact models can associate hex values with color semantics. We present Paint-Anything, which learns a shared hex-prompt interface for generation and editing through object-level color supervision. We develop a data pipeline that constructs Paint-500K from real images through object grounding, perceptual color labeling, and editing-pair synthesis. Since shadows make real-image labels only approximate colors, we supplement supervision with solid-color anchors whose pixels exactly match paired hex values; these anchors are used only at high-noise timesteps, leaving low-noise training to natural images. We also introduce ACBench to measure object-level hex color fidelity in both tasks.

*Paper*