ComfyUI Node
TS Free VRAM
Take models off the GPU at a chosen point in the graph. Passes its input through untouched, so put it on a link that is used right before a step that needs the whole card — a heavy VAE decode, an upscale, a second sampler.
TS Free VRAM
- passthrough
- model
- passthrough
CategoryTS/Utils
Inputs (2)
| Name | Type | Default | Description |
|---|---|---|---|
| passthrough | * | Anything at all — it comes back out unchanged. Its only job is to place this node in time: ComfyUI runs a node when its output is needed, so put this on a link consumed AFTER the heavy step you want to free memory for (a latent going into VAE Decode is the usual spot). On a link feeding a sampler it would run too early. | |
| modelopt | MODEL | Which model to unload. Connect the same MODEL the sampler used and only that one is taken off the card. Leave it empty to unload everything currently loaded. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| passthrough | * | Exactly what came in, unchanged. |