JD.com Open-Sources JoyAI: Real-Time Streaming Video Editing Model
fruesome · reddit · 2026-08-11
JD.com has open-sourced JoyAI-Video-Edit, a real-time, instruction-guided video editing system based on autoregressive diffusion. Unlike traditional offline batch processing, this model edits frames causally as they arrive, without waiting for the full video or requiring a predefined length.
- Core Architecture: Combines an MLLM-based condition encoder, a causal video VAE, and a 16B-parameter multimodal diffusion transformer (MMDiT).
- Acceleration: Utilizes aligned autoregressive distribution matching (ADM) distillation and bounded KV-state inference.
- Performance: Achieves an end-to-end throughput of 30.19 FPS at 720x1280 resolution, pushing video editing towards interactive streaming generation.
More from Multimodal
- Open-Source Realism LoRA for MiniMax H3: Curing the Plastic Look in AI Video — Affectionate-Map1163 · 2026-08-11
- Testing MiniMax H3's Celebrity Character Generation on RTX 5090 — JuniorEnsign · 2026-08-11
- Major Unofficial A1111 Fork Adds Native Python 3.14 and Pony Model Support — Zestyclose_Bake3680 · 2026-08-11
- MiniMax H3 Audio/Video 8-Step Turbo LoRA Surfaces on Hugging Face — ThatsALovelyShirt · 2026-08-11
- Fixing Intel Arc VRAM Issues: Developer Releases Stability Patch for ComfyUI — Valuable-Subject-274 · 2026-08-11
- Multimodal AI Workflow: Creating a Ken Block Tribute Video in Under a Day — beechinour · 2026-08-11