🚀Generate a 30-second 1080p video in just 7 seconds! We’re open-sourcing FastVideo Dreamverse: real-time vibe directing for video generation on a single NVIDIA B200 GPU with LTX-2 model LTX Repo: Blog:

Video generation should keep creators in the loop, not pull them out of it. Video generation should be fast enough for creators to try an idea and polish it immediately. Dreamverse lets users: - generate a scene - steer it with natural language - rewrite the direction - continue the video - iterate quickly Less waiting. More directing.

Dreamverse is our flagship application built on the FastVideo framework. It brings together the core components of a real-time generative video system: - Inference optimization - Streaming support - GPU resource management - Prompt rewriting - App session orchestration

Under the hood, Dreamverse uses a NVIDIA B200-optimized FastVideo generation path with: - FA4 FlashAttention - NVFP4 inference - torch.compile - continuation-aware generation - low-latency streaming Checkout to see how you can use these optimizations in your FastVideo application!

We see Dreamverse as a reference application for the next generation of interactive video tools: - Creative workflows - Editing systems - Real-time agents - Generative media applications All built around one idea: Video generation should be steerable, fast, and interactive.

Huge thanks to the FastVideo team and contributors who made this release possible. Special thanks to our collaborators and supporters across the ecosystem, including @NVIDIAAI , @ltx_model , @mbzuai , for helping push real-time video generation forward on modern GPU platforms.

@ltx_model We have reached video gen escape velocity 🚀 Meaning you could continuously watch a full ai movie being generated on the fly, efficiently amazing work 👏

@ltx_model Wow this is great! Getting this to run on consumer GPUs would be so useful.

@ltx_model This is actually impressive. Really cool to see it open-sourced too.

@ltx_model 7 seconds for 30 seconds of 1080p video is the kind of speed jump that changes behavior. Creative iteration gets weird once waiting stops mattering.

@ltx_model Hey @Prince_Canuma, please please say your thing!!

@ltx_model No more “generate and pray”, this is true interactive video gen

@ltx_model That is incredible efficiency.

@ltx_model Congrats @haoailab !

@ltx_model Wow impressive work

@ltx_model That's so amazing!

@ltx_model Damn this is fast

@ltx_model @Presidentlin

@ltx_model Sweet! Thanks for sharing

@ltx_model CC @cartyisme

@ltx_model Project looks solid 🔥 open to collab?

@ltx_model I am spinning a pipeline with this up in my spark. Can’t wait!

@ltx_model No LTX 2.3 support??

@ltx_model How?

@ltx_model 🔥🔥🔥

@ltx_model so wonderful

@ltx_model What a great easy way to generate videos, great work from Hao

@ltx_model Quality is what matters how does it fair to Seedance

@ltx_model @grok take this code and make it work on esp32

@ltx_model FastVideo Dreamverse makes real-time video generation on a single GPU actually possible

@ltx_model @haoailab The 7 seconds is time to stream first chunk while later keeps generating ? Wasn't able to generate full 30 seconds in 7 seconds on B200 !

@ltx_model 🔥⚡️

@ltx_model AMD? Rocm?

@ltx_model 30 seconds of 1080p video in 7 seconds and they open sourced it... the barrier just dropped again

@ltx_model Can you port this to Nvidia DGX Spark?