DreamSR: Towards Ultra-High-Resolution Image Super-Resolution via a Receptive-Field Enhanced Diffusion Transformer

Researchers introduced DreamSR, a diffusion-transformer super-resolution framework for ultra-high-resolution images.

It uses a dual-branch MM-ControlNet design plus receptive-field enhancement training to reduce patch mismatch and over-generation.

The approach improves fine texture synthesis and restores sharper, more faithful details in large images.

The work targets a common failure mode in diffusion-based SR, where patch-wise inference can hurt consistency.