Robust-U1

Robust-U1: Can MLLMs Self-Recover Corrupted Visual Content for Robust Understanding?

Input
Source Image and Edit Instruction
Output
Generated Result and Trace
0 1000000
1 8
1 4
0 1
CFG renormalization type
0 1
10 100
1 10
64 4006
0.1 1
Inference is powered by the bundled modeling package in this Space. Model weights are fetched from the Hugging Face Hub repo/path Jiaqi-hkust/Robust-U1 when the runtime is first initialized.