Note: All model files and custom nodes below must be downloaded via the client. If you do not have the client, Download >>>
RelayVSR is an innovative and highly efficient Video Super-Resolution (VSR) and enhancement framework built on a collaborative "Large-Small Model" architecture utilizing a Sparse Generative Relay mechanism. Moving away from the heavy overhead of processing every single frame, it strategically leverages a powerful generative foundation model to process sparse keyframes (such as every 15th frame and the final frame) as reliable temporal anchors, while smaller models and VAE decoders efficiently fill in the intermediate frames. This design delivers stunning x2 and x4 upscale details, realistic textures, and exceptional clarity at remarkable speeds, making it a powerful and seamless addition to the ComfyUI ecosystem.
| Name | Size | Target Dir |
|---|---|---|
| relayvsr_flash_vae_x2_fp32.safetensors | 127.76 MB | vae |
| relayvsr_flash_vae_x4_fp32.safetensors | 127.76 MB | vae |
| Name | Size | Target Dir |
|---|---|---|
| relayvsr_x2_lightx2v_int8_convrot.safetensors | 5.16 GB | diffusion_models |
| relayvsr_x4_int8_convrot.safetensors | 5.16 GB | diffusion_models |