Plot Points

Points as numbers are fine for a machine. For a human, you want to see them. Plot Points takes a set of normalized points and a source image, and stamps visible markers on top so you can verify the vision model actually pointed at the right thing. It's the visual half of the pack's Molmo pointing pipeline: generate text → Parse Points → this node → eyeball the result.

Inputs

• points - a POINT array from Parse Points (normalized 0–1 coordinates).
• image - the IMAGE the points refer to.
• size - marker radius in pixels, default 5.
• font_size - label text size, default 40. Set to 0 to skip labels entirely.
• color - default #0000ff (blue). Can be a comma-separated list of hex colors, which is where it gets clever: the node maps each unique label to its own color, cycling through the list. Labeled points become color-coded automatically.
• labels (optional) - a STRING list from Parse Points. Without labels, every marker is just a dot.

Output is an IMAGE with the markers drawn on - same batch, same resolution as the input.

How it works

The node matches points to labels per batch item, then draws a filled circle at (x * width, y * height) with Pillow, plus the label text above the marker. If the batch sizes don't line up it warns and trims to the smallest. It's simple, and simple is what you want from a debug-visualization node.

Where it fits

The README's Molmo multi-pointing example is the canonical use: two images, two prompts, two Parse Points runs, one Plot Points to draw both label sets - the author found a single prompt assigning two labels to two points just merges them, so two simple prompts is the working pattern. Also handy for sanity-checking any "point to the X" answer before you trust the coordinates downstream.

Install

Part of the SeanScripts pack - Manager search ComfyUI-PixtralLlamaVision, or clone the repo. Pure Pillow under the hood, so no extra dependencies beyond the pack's core ones (transformers, accelerate, bitsandbytes, torchvision). Restart after installing.

Troubleshooting

• Points land in the wrong spot - check that the coordinates are normalized 0–1. If you fed it pixel coordinates from something else, it'll silently map them wrong.
• No labels drawn - font_size of 0, or no labels input. Both skip text.
• Markers off-image - a point slightly outside 0–1 will draw off the edge; re-check your parse.

It won't change your life on its own, but the moment Molmo points at a face and you want to see the dot, this is the node.