Track synthesis, materials and catalysis as it is published.
Search papers, labs, and topics across Lattice.
SenseTime Research, HKUST MMLab
4
0
7
Packed trajectories, not sequential rollouts, unlock high-fidelity multi-turn image editing by preventing visual drift.
Current video generation benchmarks miss the forest for the trees: EvalVerse actually measures cinematic quality, not just prompt adherence.
Unified multimodal models aren't truly unified: vision and language modalities exhibit divergent entropy patterns during encoding and generation, hindering effective reasoning-based image synthesis.