← back to paper
arxiv: 2608.01150 · 2 revisions
Zellige: Moldable Sequence Placement for Mixed Image-Video DiT Training