ComfyUI Node
Isolate Input SAM3.1 (CRT)
A ComfyUI node in CRT/Utils/Isolate with 10 inputs and 2 outputs.
Isolate Input SAM3.1 (CRT)
- images
- cropped_images
- pipe
◄what_to_detectface►
◄performance_presetQuality►
◄single_itemfalse►
◄detect_chunk_size0►
◄padding512►
◄threshold0.30►
◄bbox_expansion0.75►
◄crop_smooth_alpha1.00►
◄crop_megapixels1.00►
CategoryCRT/Utils/Isolate
Inputs (10)
| Name | Type | Default | Description |
|---|---|---|---|
| images | IMAGE | — | |
| what_to_detect | STRING | face | Text prompt used for subject detection, for example 'face' or 'person'. |
| performance_preset | COMBO | Quality | Fast = quickest detection, Balanced = default tradeoff, Quality = highest available quality in the current custom-node path and slower processing. The SAM path still uses a model-safe square working resolution. |
| single_item | BOOLEAN | false | Limit detection to one item per frame. Enable this when you only want the main face/subject. |
| detect_chunk_size | INT | 00–4096 | Detection batch size. 0 = process the whole batch at once, 1 = per-image processing, 2+ = fixed chunk size. |
| padding | INT | 5120–2048 | Extra border added around the input batch before detection so subjects near the edges are not cropped too tightly. |
| threshold | FLOAT | 0.300–1 | Detection confidence threshold. Lower values find more masks, higher values are stricter. |
| bbox_expansion | FLOAT | 0.750.5–4 | Expands the detected crop box around the subject. Higher values give a looser crop. |
| crop_smooth_alpha | FLOAT | 1.000–1 | Temporal smoothing for crop position. Lower = steadier crop, higher = follows motion more closely. |
| crop_megapixels | FLOAT | 1.000–8 | Target resolution for the cropped output images. 0 disables crop rescaling. |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| cropped_images | IMAGE | — |
| pipe | CRT_ISOLATE_PIPE | — |