JoyAI-Video-Edit: Real-Time Open-Ended Video Editing with Autoregressive Diffusion
JoyAI says its 16B-parameter editor can process 720p video at about 30 FPS on one Nvidia B200 GPU.
The system is built for streaming edits, generating causally without future frames or a fixed video length. It uses chunk-wise autoregressive adaptation plus two distillation methods meant to preserve source fidelity and limit temporal drift. The authors report that it beats existing streaming editors and stays competitive with strong offline systems on short and long videos. HF Daily Papers' note
The system is built for streaming edits, generating causally without future frames or a fixed video length. It uses chunk-wise autoregressive adaptation plus two distillation methods meant to preserve source fidelity and limit temporal drift. The authors report that it beats existing streaming editors and stays competitive with strong offline systems on short and long videos. HF Daily Papers' note
score 6