ComfyUI Node
YG Clean VRAM
A ComfyUI node in YG_Nodes with 9 inputs and 1 output.
YG Clean VRAM
- passthrough
- passthrough
◄unload_modelstrue►
◄aggressivetrue►
◄max_passes3►
◄target_free_mb0►
◄trigger_every_runtrue►
◄clean_all_serversfalse►
◄dashboard_urlhttp://127.0.0.1:8194►
◄dashboard_token►
CategoryYG_Nodes
Inputs (9)
| Name | Type | Default | Description |
|---|---|---|---|
| unload_models | BOOLEAN | true | Unload all loaded models from VRAM. Disable to keep the model cache (faster next run, more VRAM used). |
| aggressive | BOOLEAN | true | Adds cuda.ipc_collect + synchronize after empty_cache. |
| max_passes | INT | 31–10 | Number of clean-then-measure passes. Stops early once VRAM stops dropping. |
| target_free_mb | INT | 00–200000 | Optional: keep cleaning until GPU 'used' drops to this many MB or below. 0 disables. |
| trigger_every_run | BOOLEAN | true | Re-execute on every workflow run (no caching). |
| clean_all_servers | BOOLEAN | false | Also clean VRAM on ALL ComfyUI servers via the dashboard (or by POSTing /free to each port). Use this if other ComfyUI instances on the same GPU are holding memory. |
| dashboard_url | STRING | http://127.0.0.1:8194 | Server Status Dashboard base URL (e.g. http://127.0.0.1:8194). |
| dashboard_token | STRING | Dashboard SECRET_TOKEN. | |
| passthroughopt | * | Optional value forwarded to the output. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| passthrough | * | — |