ComfyUI Node
SeedVR2 Video Path Auto Configurator
Detects the selected GPU and configures model precision, attention, batching, tiling, model residency, tensor offload, and direct-video chunking. Connect all three configuration outputs to the SeedVR2 Video Path Upscaler.
SeedVR2 Video Path Auto Configurator
- video
- dit
- vae
- auto_settings
- report
◄deviceauto►
◄profilebalanced►
◄target_resolution1080►
◄reserve_vram_gb4.0►
◄batch_size_override0►
◄enable_torch_compilefalse►
CategorySeedVR2 Video Path
Inputs (7)
| Name | Type | Default | Description |
|---|---|---|---|
| device | COMBO | auto | 1 options: auto |
| profile | COMBO | balanced | 3 options: balanced, maximum_quality, maximum_throughput |
| target_resolution | INT | 108016–4320 | Desired output short-edge resolution; hardware settings adapt around it. |
| reserve_vram_gb | FLOAT | 4.00–64 | VRAM kept free for ComfyUI, previews, and other nodes. |
| batch_size_overrideopt | INT | 00–16381 | Override the recommended temporal batch size. 0 keeps automatic sizing. Non-zero values must follow 4n+1: 1, 5, 9, 13, ... |
| enable_torch_compileopt | BOOLEAN | false | Enable only for repeated runs; the first run has significant compilation cost. |
| videoopt | VIDEO | Optional native VIDEO used to avoid oversized batches/chunks on short clips. |
Outputs (4)
| Name | Type | Description |
|---|---|---|
| dit | SEEDVR2_DIT | — |
| vae | SEEDVR2_VAE | — |
| auto_settings | SEEDVR2_AUTO_SETTINGS | — |
| report | STRING | — |