KAIST AI / Bryan Sangwoo Kim et al. · Image Editing

Chain-of-Zoom (CoZ)

CoZ is a model-agnostic framework that enables extreme super-resolution (16x-256x magnification) by decomposing upscaling into an autoregressive chain of intermediate scales with VLM-generated multi-scale-aware prompts, fine-tuned using GRPO for human preference alignment. A standard 4x diffusion SR model wrapped in CoZ achieves beyond 256x enlargement without additional training.