Image_expand_canvase_visual

The outpainting trick that's taken over from painting an extension mask by hand is almost embarrassingly simple: put your image on a bigger canvas filled with a flat color, then either mask that new region for a classic inpaint pass, or - with a modern instruction-editing model - just tell it in the prompt to "fill in what's behind the padding." One workflow that circulates for Flux Klein does exactly this: pad each side with 200px of flat red, then prompt "remove the red paddings on the side and show what's behind them." Either route needs a node that builds the padded canvas and hands back a mask marking exactly what's new - that's this node, with drag handles instead of typed-in numbers.

How it works

width/height (both default 2048) set the target canvas size. expand_state - a small JSON object tracking left/right/top/bottom - is driven by dragging handles directly on the node rather than typing four numbers. constant_color (white/black/red/gray/edge) fills the new area; "edge" stretches the border pixels instead of a flat fill, which avoids an obvious seam if you're not deliberately using a flat color as an instruction-model cue.

The inputs and outputs that matter

• width/height (default 2048 each) - target canvas dimensions.
• constant_color (default "black") - fill for the newly added area, or "edge" to extend the border pixels instead.
• expand_state - the drag-driven left/right/top/bottom offsets, stored as JSON.
• image (optional) - what gets placed on the new canvas.
• Output: image (the expanded canvas) and pad_mask (marks the newly added region - feed this straight into whatever inpaint or outpaint sampling step follows).

is_output_node is true, so it previews its own result inline - part of the same 2026-03-22 visualization batch as the pack's other _visual nodes.

How to install it

Via ComfyUI Manager: search ComfyUI-Apt_Preset, install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/cardenluo/ComfyUI-Apt_Preset.git

Windows: install.bat. Linux/Mac: pip install -r ComfyUI-Apt_Preset/requirements.txt in your venv. Restart. Nothing to download for canvas math - the pack's heavier dependencies belong to other parts of this ~300-star, actively maintained pack.

Common issues & troubleshooting

Output just repeats the flat pad color instead of generating into it. This node only builds the canvas and the mask - it doesn't generate anything itself. Check that pad_mask is actually wired into your sampling step; a disconnected mask means nothing downstream knows where the new region is.

Using a flat color for an instruction-model prompt trick and it's not cooperating. An uncommon color works better than white or black, which a model might read as an intentional part of the image rather than filler to remove - that's the logic behind the flat-red-padding trick mentioned above. If you're doing a classic masked inpaint instead, "edge" is usually the safer default since it gives the model a plausible-looking extension to build on rather than an obvious hard color.

Drag handles not registering. expand_state only updates from actual interaction with the node's widget - changing width/height alone won't move the left/right/top/bottom offsets; you need to drag the handles themselves for those to change.