Search papers, labs, and topics across Lattice.
1
0
3
Autoregressive video models can now generate 4-minute videos without retraining, thanks to a clever inference-time hack that fixes positional embedding bias and injects dynamic priors.
