Source code</>
UNREEL
Experimental
UNREEL
An open-source, self-hosted AI streaming service that writes scenes and renders MiniMax H3 Max Turbo video through fal as a channel plays.
Open source / Autoregressive streaming video research
Self Forcing
Research code that trains video diffusion models on their own autoregressive rollouts and provides streaming-generation inference.
View repository ↗Source code</>
Self Forcing
The unlock
Why this one is on the map.
A building block for continuous AI video; a channel still needs playback, input handling and broadcast plumbing.
The clearest product signal is “Train for the way a stream unfolds”. That interaction model—not a generic AI feature list—is the reason to compare it with the rest of this directory.
This is source code, not a hosted channel. You need to inspect, configure, and deploy it with your own keys and runtime before anything can broadcast.
How to read it
What happens when you open it.
Open the official destination and look for the current research model · streaming inference · gpu workflow. Controls, access and live activity can change after this editorial check.
Best for: Autoregressive streaming video research
Access note: Read the repository setup and license before running. Hardware, model downloads and API access vary by project.
Sources & status
Keep the link, keep the caveat.
2026-09-05: Official README reviewed, including inference and minimum-memory instructions. No local model run or end-to-end live application tested.
Explore similar projects
Suggest a correction ↗Source code</>
Daydream Scope
Experimental
Daydream Scope
A workbench for running real-time generative video pipelines, with model selection, composable processing and WebRTC, NDI, Spout or Syphon output paths.
Source code</>
StreamDiffusionV2
Experimental
StreamDiffusionV2
An interactive video-diffusion system with released checkpoints, configurable inference and support for different GPU arrangements.