CV Plot 2D (cv2.plot)

Charts inside ComfyUI are usually either interactive widgets (great for reading, useless for output) or a bunch of drawing nodes. CV Plot 2D takes the third path: it renders a plot as an ordinary IMAGE using cv2.plot.Plot2d, OpenCV's own plotting module. Because it's just an image, it composites, diffs, saves and stacks like anything else - which is exactly what you want when the plot is going into something: a report sheet, a video frame, a PNG that lands next to the data it describes.

It draws one series, with no legend and no labels. This is not the ECharts family - if you need several series, category names, tooltips or a threshold line, the pack's CV Chart Series / CV Chart Scatter / CV Chart Matrix nodes are the interactive, fancier option. Pick by destination: image out → this; inspect in the UI → those.

What goes in

values is the series. A 1-D array (or (N,1) / (1,N)) gets plotted against its sample index; an (N,2) array is read as x,y pairs. x_values (optional) overrides the x axis, ignored when values is already pairs. Larger values are drawn upward by default.

Then the layout: width (default 640) and height (default 400). Watch the clamps - cv2.plot itself clamps width up to 400 and height up to 300, so asking for a 200-wide plot gets you a 400-wide one. That's the module, not the node.

draw picks line (connects consecutive samples) or points only (marks them without segments). Points-only is the honest choice when the samples are independent measurements rather than a signal - joining them implies continuity you don't have.

Cosmetics are all optional widgets: line_width, show_grid, grid_lines, and the colors as BGR tuples (line_color (0, 255, 0), background_color (30, 30, 30), axis_color, grid_color, text_color). BGR, not RGB - the pack is consistent about this, but it's the kind of thing you'll get wrong once.

Two more that earn their place:

• value_range - auto (data range) lets Plot2d fit the axes, which is convenient and hides how big the values actually are; manual pins the axes with min_x / max_x / min_y / max_y. Manual is how you compare two runs side by side, and it's the reason those four inputs exist.
• print_point_index - annotates ONE sample's value onto the plot. -1 is none. Perfect for calling out the winner in a comparison without a second annotation pass.
• y_down (default false) flips the axis so larger values go down. Worth knowing: this is the opposite of raw cv2.plot, whose own default draws downward - the node corrects it, so the default here matches every other plot in the pack.

One output: plot, an IMAGE (BGR converted to ComfyUI RGB) ready for Save Image or compositing.

Install

Manager → search ComfyUI CV, or:

cd ComfyUI/custom_nodes
git clone https://github.com/bmad4ever/comfyui_cv
cd comfyui_cv && pip install "opencv-contrib-python-headless~=5.0.0.93"

Restart. Python ≥ 3.12, V3 node API. Plot2d is a cv2 class that only accepts float64, which the node handles for you - feed it whatever dtype you have.

Common issues

• Blank or empty plot - either the input was empty (which renders an empty plot rather than failing) or the values are all identical and the auto range collapsed. Use manual range to check.
• Your requested size was ignored - the 400/300 clamps above.
• The colors are wrong - BGR tuples. Swap the first and third numbers.
• Multiple series came out only once - one series per node. Use CV Chart Series if you need more, or stack images.
• Values went the wrong way up - y_down is off by default specifically so this node doesn't inherit cv2.plot's downward convention. If you turned it on for image-coordinate data, that's why.

Pack disclaimer as usual: LLM-assisted, personal, not production-grade, no promised updates. The nice thing about a plot is that a wrong one usually looks wrong.