| --- |
| license: unlicense |
| tags: |
| - flux2 |
| - flux |
| - video-to-video |
| - stream-to-stream |
| - diffusion |
| - flow-matching |
| - rectified-flow |
| - generative-models |
| - image-generation |
| - pytorch |
| library_name: pytorch |
| pipeline_tag: video-to-video |
| --- |
| # Stream Refractor |
|
|
| **Stream Refractor** is an experimental *"Stream2Stream"* model that edits input video stream in real-time according to *edit refernce* produced by large image editing model (e.g. Flux.2-klein) |
|
|
| It is a **Rectified Flow** model with **0.6B** parameters that generates each latent block of frames in a few diffusion steps. |
|
|
| ### Inference code: [https://github.com/tensorforger/StreamRefractor](https://github.com/tensorforger/StreamRefractor) |
