Bria Expand Image

You've got a great image that's the wrong shape - a portrait you need as a 16:9 header, a square product shot that has to become a vertical ad, a photo with the subject jammed against the edge. Local outpainting is a chore: load an inpainting model, fiddle with a denoise setting, babysit the seams. Bria Expand Image is the "just make it bigger" button instead. You hand it an image and a target shape, and Bria's hosted model generates whatever's missing beyond the borders.

Here's the part people miss on first glance: this is a partner node, not a local one. There's no checkpoint to download, no VRAM touched, no prompt-engineering religion. It uploads your image to Bria's API - routed through your Comfy account - and bills your prepaid credits per call. It's one of the new Bria family that landed in ComfyUI core in August 2026 (alongside GenFill, Eraser, Remove Background and the video nodes), so you need a recent ComfyUI build and nothing else. No model files go anywhere on your disk.

The one input that matters: expand_mode

Everything else is gravy. Pick how the expanded image should be shaped:

• Preset ratios - 1:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9. Pick the target, done.
• custom_ratio - reveals ratio_width and ratio_height (1–100 each) for odd shapes. Gotcha: Bria only accepts a width-to-height ratio between 0.5 and 3.0, so anything taller than 1:2 will error with a message telling you to switch modes.
• manual - the power mode. You get canvas_width/canvas_height (64–5000) plus image_width/image_height and image_x/image_y, the position of your image's top-left corner on that canvas. It's the only mode that can reach a canvas taller than 1:2, and since the image may sit partly off the canvas, you can also use it to crop while you expand. This is where you go when the presets don't fit your layout.

prompt is optional and multiline. Leave it empty and Bria writes its own description from the image - and hands it back to you on the second output, which is genuinely handy if you want to reuse it downstream. There's a negative_prompt, a seed (42 by default, wired to "control after generate" like the rest of ComfyUI), and a moderation combo that expands into prompt/visual-input/visual-output moderation toggles. Default off; flip them on if you're feeding public-facing content through it.

What comes out

Two outputs: the expanded IMAGE, and the prompt string that was actually used. Wire the image into a Save Image and the prompt into whatever text node wants it. Mechanically the node uploads your image, polls Bria's status endpoint until the result is ready, and downloads it back - so expect a few seconds of wait, and the image does leave your machine. Don't use it for anything you can't send to a third party.

Where people get burned

The tall-ratio error is the #1 stumble - you set 9:21 in custom mode and it refuses; that's the 0.5–3.0 limit doing its job, not a bug, and manual mode is the fix. The other failure class is account-shaped: partner nodes need you logged into a Comfy account with a credit balance above zero. No login, no credits, no network to comfy.org and the node just fails on the call. And if the Bria nodes are cluttering your node search and you'll never touch them, the community's answer is the --disable-api-nodes launch flag - they vanish from the menu until the next update re-adds them. Pricing is per call in prepaid credits and moves around, so check the badge on the node rather than trusting a number you read anywhere.