ComfyUI Node
YOLOE-26 Load Model
Load a YOLOE-26 open-vocabulary segmentation model from the ComfyUI model directories, downloading official weights on demand when auto_download is enabled. Note: the first text-prompt inference downloads the MobileCLIP text encoder (~250 MB) through Ultralytics and requires network access.
YOLOE-26 Load Model
- model
◄model_nameyoloe-26s-seg.pt►
◄deviceauto►
◄auto_downloadtrue►
◄offload_to_cpufalse►
CategoryYOLOE26
Inputs (4)
| Name | Type | Default | Description |
|---|---|---|---|
| model_name | COMBO | yoloe-26s-seg.pt | YOLOE-26 model file name. Official models are downloaded automatically when auto_download is enabled; other entries are .pt files found in your ComfyUI model directories (press 'r' in ComfyUI to refresh the list after adding files). |
| deviceopt | COMBO | auto | Inference device. 'auto' lets Ultralytics choose. |
| auto_downloadopt | BOOLEAN | true | True (recommended) = automatically download the model if not found locally. False = local-only; the model must already exist in a supported ComfyUI model directory. |
| offload_to_cpuopt | BOOLEAN | false | Move the model to CPU after each segmentation node finishes to free GPU memory for diffusion models. Slightly slower because weights are transferred back to the GPU on the next run. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| model | YOLOE_MODEL | — |