Outpaint or extend an image

Base64-encoded input (reference) image or HTTP(S) image URL.

Target output width.

Target output height.

If True, crop the input image to the canvas bounds when it extends beyond the edges. If False, an error is raised instead.

Tolerance level for input and output moderation. Between 0 and 5, 0 being most strict, 5 being least strict. If you wish to increase the safety_tolerance beyond these levels please reach out to the team.

Experimental: optional text guidance for the outpainted region. The model may not strictly follow this prompt; the visual content of the input image is the primary signal. Leave unset for default behavior.

Left offset (px) of the reference image's top-left corner on the output canvas. Negative values are allowed. None = center horizontally.

Top offset (px) of the reference image's top-left corner on the output canvas. Negative values are allowed. None = center vertically.

Quality/speed trade-off. 'high' (default): highest-fidelity results, recommended whenever fine detail, prompt adherence, or consistency with complex content in the source image matters; slower. 'fast': significantly faster and well-suited for naturally extending most scenes (landscapes, backgrounds, textures, products); may produce lower fidelity in the extended region than 'high'.

Skip the image-aware prompt upsampler for lower latency. The prompt (or a default extension prompt) is sent directly to the generation model; quality in the extended region may be lower for scenes that benefit from semantic guidance. Applies to both modes.