ComfyUI Node
FlashVSR+ Tiled Model Loader
A ComfyUI node in FlashVSR+ Tiled with 7 inputs and 2 outputs.
FlashVSR+ Tiled Model Loader
- model
- model_info
◄mode▾►
◄precision▾►
◄attention▾►
◄decoder▾►
◄auto_downloadtrue►
◄keep_loadedfalse►
◄model_directory►
CategoryFlashVSR+ Tiled
Inputs (7)
| Name | Type | Default | Description |
|---|---|---|---|
| mode | COMBO | Choose the FlashVSR+ model variant. Turbo prioritizes speed; Base and V1 use their corresponding official weights. | |
| precision | COMBO | Choose DiT weight precision. INT8 ConvRot reduces weight memory; BF16 or FP16 use floating-point weights. | |
| attention | COMBO | Select the attention backend. Auto picks an available supported backend; SDPA is the portable fallback. | |
| decoder | COMBO | Choose the output decoder. Tiny supports incremental long-video decoding; Wan VAE decodes the full clip. | |
| auto_download | BOOLEAN | true | Download missing model files automatically into the configured FlashVSR+ model folder. |
| keep_loaded | BOOLEAN | false | Keep weights in CPU RAM between requests. GPU memory is released after each request. |
| model_directory | STRING | Optional model root; blank discovers models/flashvsr_plus and the bundled ckpts folder. |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| model | FLASHVSR_PLUS_MODEL | — |
| model_info | STRING | — |