← back to paper
arxiv: 2605.14876 · 2 revisions
Unlocking Complex Visual Generation via Closed-Loop Verified Reasoning