RINO: Unifies Vision Tasks as RGB Editing

Timing Yang · hf · 2026-07-15

This work introduces RINO, which unifies various visual information into RGB image representations and reformats vision tasks into an RGB-to-RGB image editing problem.

Core Concept

Results

The authors hope this will inspire unified vision-language systems, allowing different visual tasks to be expressed and solved using a shared "visual language." The code is open-source.

Original post →

More from Multimodal

Multimodal channel →