Outpainting

The photo’s composition is great, but the frame is too tight — you want more sky, more ground. That’s outpainting (canvas expansion). It’s similar to inpainting: both involve “what can change vs. what can’t” — the key difference is that the mask is usually generated for you, marking the newly added blank region around the edges.

The official example workflow is at ComfyUI Outpainting Workflow.

How It Relates to Inpainting

A rough way to think about it: inpainting digs a hole in the middle and fills it; outpainting adds blank paper around the edges and fills that.
So in ComfyUI, some outpainting workflows are built on the inpaint pipeline under the hood — they just swap in “pad the edges first, then generate a matching mask” nodes at the front.

What You Need

Model: many examples use a checkpoint trained for inpainting (the official tutorial cites something like 512-inpainting-ema.safetensors), placed in models/checkpoints. Use whatever your workflow specifies.
Input image: the one you want to expand.
Workflow: the official style is to distribute a workflow image with embedded JSON metadata — drag it into ComfyUI to load it; or use Workflows → Open.

What `Pad Image for Outpainting` Does

Think of it as: adding blank padding around the original image, then generating a mask that marks “which part is original, which part is new padding.” The encoder and sampler use this to know where to “continue drawing” without disturbing the original center.

Parameters at a glance (names are enough — details vary by UI version):

Parameter	Plain English
left / top / right / bottom	How many pixels to expand in each direction
feathering	How soft the transition is: larger = less of a “hard cut” at the seam

Output is typically two things: the padded image and the accompanying mask.

What to Click When Running

Load Checkpoint: select the right model file.
Load Image: upload your source image.
Queue or Ctrl / Cmd + Enter to run.

Then experiment: how much to expand, how to write prompts, how many seeds to try — same as text-to-image: get the direction right, then refine parameters.

One-Line Summary

Outpainting = first “make the canvas bigger,” then use a mask to tell the AI “fill in the new region.” Understand that, and you’ll never get lost reading anyone else’s outpainting workflow.