SAMLoader (Pipe)

If you do any detail-fixing - FaceDetailer-style detect-crop-refine passes - you know the loop: a detector finds a region, SAM (Segment Anything Model) turns that rough box into a clean mask, and a fresh sampling pass repaints it. easy samLoaderPipe loads the SAM half of that and bundles it, plus its segmentation settings, into a single pipe you hand to Easy-Use's detailer nodes.

This is Easy-Use wrapping the SAMLoader concept from ltdrdata's Impact Pack - the pack that made automatic detail refinement standard practice in ComfyUI. Easy-Use re-exposes it in "pipe" form so the loader and all its knobs travel down one wire instead of cluttering your detailer node.

How it works

SAM takes an image plus a hint (a point, a box, a rough mask) and produces a precise segmentation mask. This node loads the SAM checkpoint and pre-packs the hint strategy and threshold settings into a sam_pipe. Downstream, the detailer feeds it the region a detector found, and SAM tightens it into a mask worth repainting.

The inputs and outputs that matter

Most people set two things and leave the rest:

• model_name - the SAM checkpoint (e.g. sam_vit_b_01ec64.pth). This has to be downloaded and present or the dropdown is empty.
• sam_detection_hint - how SAM is told where to look: center-1, rect-4, mask-area, and so on. center-1 (a single center point) is a fine default for tidy single-subject regions.

The finer dials - sam_threshold (default 0.93, how confident a pixel must be to count), sam_dilation (grow/shrink the mask edge), sam_bbox_expansion, device_mode (AUTO / Prefer GPU / CPU) - are there when you need them, but the defaults are sensible. Output is a single sam_pipe (PIPE_LINE); wire it into an Easy-Use detailer.

How to install it

ComfyUI Manager: search ComfyUI-Easy-Use, install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/yolain/ComfyUI-Easy-Use

then install.bat / pip install -r requirements.txt, restart. You need a SAM model file - download a SAM checkpoint (the ViT-B one is the small, fast default) into ComfyUI's models/sams folder so model_name can find it.

Common issues & troubleshooting

model_name dropdown is empty. No SAM checkpoint installed, or it's in the wrong folder. Put a .pth SAM file in models/sams and restart.

Masks are ragged or grab too much. Nudge sam_threshold up for a tighter, more conservative mask, or use sam_dilation to shrink (negative) or pad (positive) the edge. If SAM keeps segmenting the wrong thing, switch the sam_detection_hint - center-1 assumes the subject is central; mask-area or rect-4 behave differently.

Slow on CPU. device_mode defaults to AUTO; force Prefer GPU if SAM is running on CPU and dragging your detailer pass.

Bigger SAM models are overkill here. ViT-B is plenty for face/hand detailing and much lighter than ViT-H. Unless you're segmenting genuinely tricky scenes, don't pay the memory cost of the huge checkpoints.