ComfyUI Node
Configure FlashVSR Upscaling
Injects FlashVSR LQ conditioning into a stock Wan model. Compatible TensorWise INT8 ConvRot checkpoints automatically reuse one INT8 activation preparation across Q/K/V during streaming. Q/K/V remain separate ComfyUI-managed weights, so Dynamic VRAM can lease them sequentially; non-compatible checkpoints use stock projections.
Configure FlashVSR Upscaling
- model
- lq_projector
- video
- MODEL
- FLASHVSR_RUNTIME
◄conditioning_strength1.00►
CategoryFlashVSR
Inputs (4)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | — | |
| lq_projector | FLASHVSR_LQ | — | |
| video | FLASHVSR_VIDEO | — | |
| conditioning_strength | FLOAT | 1.000–2 | Multiplier for resized-video conditioning injected at Wan block 0. 1.0 matches the checkpoint. Lower values give the model more freedom; higher values follow the input more strongly and may preserve its artifacts. |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| MODEL | MODEL | — |
| FLASHVSR_RUNTIME | FLASHVSR_RUNTIME | — |