RelayVSR: Large-Small Model Collaboration for Efficient Real-World Video Super-Resolution
Paper • 2609.37850 • Published
RelayVSR
RelayVSR performs streaming 4× video super-resolution by using a large generative model to produce sparse references and a lightweight VSR network to process the video. See the GitHub repository for inference instructions and the paper for training details.
License
The checkpoints in this repository are distributed under the Apache License 2.0. See LICENSE for the full text.
| File | Provenance |
|---|---|
wan.safetensors |
The reference generator, adapted from Wan2.2-TI2V-5B for video super-resolution. The upstream Wan2.2 model is licensed under Apache-2.0. This checkpoint also includes the prompt embeddings used for inference. |
flash.safetensors |
RelayVSR's lightweight VSR network checkpoint. |
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support