FLUX Outpainting: 4 MP canvas cap vs Sume's image size options

FLUX Outpainting takes an image and a target width and height, with the canvas area capped at 4 megapixels, and fills the new area around the placed image. Sume documents no outpainting mode. What it does document is custom sizes through image_size on the GPT ids, a list of normalized aspect ratios, and aspect_ratio: "auto" on edits.

BFL facts are from its FLUX Outpainting page and release notes; Sume facts from the Image API docs, read 2026-09-30.

What are FLUX Outpainting's limits?

The endpoint is POST /v1/flux-tools/outpainting-v1. width and height are at least 64 and their product must not exceed 4 MP (4,194,304 pixels); above that it returns 422. reference_offset_x and reference_offset_y place the reference image's top-left corner on the canvas, negative values allowed, and leaving them unset centers it. mode is high (default) or fast.

What canvas sizes does Sume accept?

For ChatGPT Image 2.5, image_size accepts named presets, auto, or custom pixels: both edges multiples of 16, a maximum edge of 3840, an aspect ratio of at most 3:1, and 655,360 to 8,294,400 pixels in total. The normalized aspect_ratio list includes wide and tall shapes such as 21:9, 9:21, 2:1 and 1:2. A model only accepts what its catalog descriptors list; unlisted parameters are rejected.

How do I get the extended look on Sume?

Pass your image as a reference (up to 16 on ChatGPT Image 2.5), ask in the prompt for the scene to continue past its edges, and set the wider canvas with image_size or aspect_ratio. There is no offset control, so the source image is not pinned to a position you choose; check the result. See custom sizes and ratios.

What about aspect_ratio auto?

On edit and image-to-image calls the docs say to prefer aspect_ratio: "auto" so the output matches the reference; omitting the field is not the same as auto. Use auto when you want the same shape back, and a listed ratio when you want a new one. To change shape only, see the aspect ratio changer.