SAM3 Create Point

The no-mouse way to make a point prompt. Instead of clicking on an image like the Point Collector does, you type an x/y coordinate and a flag for whether it's a foreground or background hint, and this node emits one point prompt. Small node, single job - but it's the building block for anything scripted or repeatable, because a typed coordinate does the same thing every run while a mouse click doesn't.

Where you'd actually use it: driving points from other nodes. If a coordinate is computed upstream - a face detector, a math node, an automated pass - you can't click that in. You feed the numbers here. On its own one point is rarely enough, so this almost always pairs with SAM3 Combine Points, which merges several of these into the SAM3_POINTS_PROMPT that a segmentation node consumes.

How it works

It takes a normalized coordinate - values from 0 to 1, where (0,0) is the top-left corner and (1,1) is the bottom-right - and wraps it as a single SAM3_POINT_PROMPT. Normalized means it's resolution-independent: the same (0.5, 0.5) hits the center whether the image is 512px or 4K. The foreground flag decides whether SAM3 reads the point as "include here" or "exclude here."

The inputs and outputs that matter

All three inputs are the whole node:

• x (default 0.5, range 0–1) - horizontal position, left to right, as a fraction of width.
• y (default 0.5, range 0–1) - vertical position, top to bottom, as a fraction of height.
• is_foreground (default on) - on = a positive/include point; off = a negative/exclude point.

Output is point_prompt (SAM3_POINT_PROMPT) - a single point. Route it into SAM3 Combine Points to bundle it with others into the SAM3_POINTS_PROMPT that Point Segmentation or Video Segmentation actually takes.

How to install it

Comes with the pack. Manager: search SAM3, install the highest version. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/PozzettiAndrea/ComfyUI-SAM3.git
cd ComfyUI-SAM3
pip install -r requirements.txt --upgrade
python install.py

then restart. The sam3.pt model in ComfyUI/models/sam3/ is needed by the segmentation nodes; this little builder itself just makes a coordinate. The README notes an experimental comfy-env/pixi one-click install.

Common issues & troubleshooting

The point lands in the wrong spot. These are normalized 0–1 fractions, not pixels. If you typed 256 expecting a pixel, that's way out of range - divide by the image dimension. (0.5, 0.5) is dead center.

My single point does nothing useful. One point is a weak hint, and this node's output is a SAM3_POINT_PROMPT (singular) that most segmentation nodes won't take directly - they want the plural SAM3_POINTS_PROMPT. Run it through SAM3 Combine Points first, ideally with a couple more points.

x and y feel flipped. x is horizontal (width), y is vertical (height), origin top-left. If your subject's on the right, x is high; if it's near the bottom, y is high.

When to skip this node entirely. If you're segmenting by hand, the Point Collector's click canvas is faster and less error-prone. Create Point earns its place only when the coordinate comes from somewhere other than your mouse.