Open-weights 16B autoregressive diffusion model from JD for real-time, instruction-guided editing of streaming video.
Updated EU status checked Last shown
Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
JoyAI-Video-Edit is a real-time, instruction-guided video editing system from JD. Given a live camera stream or an uploaded video and a natural-language instruction, it edits frames causally as they arrive, without needing the full video, a predefined length, or future frames. It combines an MLLM-based condition encoder, a causal video VAE and a 16B-parameter multimodal diffusion transformer, trained as an autoregressive diffusion editor.
The project reports 30 FPS end-to-end at 720 × 1248 in its deployment benchmark, using distillation (aligned autoregressive distribution matching), bounded KV-state inference and deployment scheduling. Supported edits include subject add/remove/replace, local edits, background changes, style transfer, motion changes and reference-image-guided editing (RV2V), such as putting clothing from a reference image on a person in the video. Version 1.1 (November 2026) improves reference-guided character replacement while preserving source motion.
Code and weights are released under Apache-2.0 on GitHub and Hugging Face, with deployment code, example training scripts and the LongV2VBench evaluation dataset. A browser demo on Hugging Face Spaces accepts a webcam stream. Real-time performance depends on hardware: the README cites 840 × 480 at 24 FPS on a single RTX 5090 (32 GB) or RTX PRO 6000. A Diffusers pipeline is listed as not yet available.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Open-source prompt-based video editor, a 16B multimodal autoregressive diffusion transformer generating 720p at 30+ fps with around one second latency. Quality matches Bernini R and Kling 3 Omni on an editing benchmark; Apache 2 license, 32.5 GB weights.
Closest neighbours by meaning, not just by category.
Model
Research framework for real-time, causal streaming video editing with diffusion, running at about 12.66 FPS.
Model · ByteDance
ByteDance's open-weights video generation and editing model pairing a Qwen2.5-VL planner with a Wan2.2 renderer.
Model · inclusionAI
Open-weights 6B unified text-to-image and instruction-based image editing model family from inclusionAI.
Model · Pruna AI
Open-weight image generation and editing model from Pruna AI, distilled from MiniMax's H3 video model to run in 8 steps.
Model · Microsoft
Microsoft's open-weights 4B model for image generation and editing, with a 4-step turbo variant.
Model · ByteDance
ByteDance's open-weights one-step diffusion transformer for restoring and upscaling video and images.
