实时视频模型
Decart MirageLSD
新增“Live-stream diffusion” — restyle a live feed (camera, game, call, screen) frame-by-frame from a text prompt at near-zero latency.
decart.ai · 收录≠背书 · 2026-09-03 复核
- 运营方
- Decart
- 模型
- Causal autoregressive live diffusion (LSD)
- 延迟
- “Under 40ms per frame” (vendor-reported)
- 输出
- Live video-to-video restyle, ~24–30fps
- 开放情况
- API via Crusoe Cloud partnership
什么是 Decart MirageLSD?
Decart's MirageLSD is a causal autoregressive diffusion model that transforms an incoming video stream in real time via text prompt, without the drift that broke earlier streaming models after ~20–30 seconds. It runs on Hopper-class GPUs with custom CUDA kernels and is offered through a Crusoe Cloud partnership; latency figures are vendor-reported ('under 40ms per frame') and not yet independently benchmarked. Karpathy is an angel investor and has called it 'real-time magic'.
Decart MirageLSD 能做什么
- Causal autoregressive diffusion that restyles a live input feed frame-by-frame from a text prompt
- Avoids the error-accumulation collapse that killed earlier streaming models after ~20–30 seconds
- Runs on Hopper-class GPUs with custom CUDA kernels
- Available as an API through a Crusoe Cloud partnership
- Backed by notable angels including Andrej Karpathy, who has called it “real-time magic”
Decart MirageLSD 适合谁?
- Live video-to-video effects on a camera, game, call or screen feed
- Building realtime creative overlays without clip generation
- Studying causal live-diffusion as an alternative to autoregressive streaming models
Decart MirageLSD 的优点与不足
优点
- First-mover framing for live-stream diffusion (vendor-positioned)
- Real-time restyle opens interactive, steerable broadcast use cases
- Meaningful backing and ecosystem partnerships
不足
- All latency numbers (“under 40ms per frame”) are vendor-reported
- Frame-rate claims vary between sources (24 vs 30 fps)
- Access is API-based via partners — no open weights or local run
别人怎么说
摘自公开报道的原文引用——收集于 2026-09-03。
“The first real-time, zero-latency generative AI video model.”
常见问题
What is Decart MirageLSD?
Decart’s “live-stream diffusion” model that restyles any incoming video feed in real time from a text prompt.
How is it different from regular video models?
It transforms a live input stream frame-by-frame rather than generating a clip from scratch — a video-to-video restyle built for liveness.
How fast is MirageLSD?
Decart reports under 40ms per frame; the figure is vendor-reported and not yet independently benchmarked.
MiniMax H3
开源权重 33B 视频模型(Hailuo 3.0)——15 秒带原生立体声片段;这波浪潮的地基。huggingface.co
fal H3 Max Live
fal 实时调优的 H3 Max——5 秒 768p 视频不到 3 秒生成,快于播放。fal.ai
Krea Realtime 14B
Open 14B autoregressive model for real-time, long-form streaming video — distilled from Wan 2.1, ~1s to first frame.krea.ai
Lightricks LTX Video
Open latent-diffusion video model that generates clips faster than real time — the engine several chat-driven live loops run on.github.com