AI Outpainting extends an image's canvas in any direction — up, down, left, or right — and fills the new area with AI-generated content that continues the existing scene seamlessly. Common uses include converting portrait images to landscape, adding breathing room around a subject, or stretching a landscape into a wider panorama without re-shooting.
How much you expand per step affects result quality
The minimum expansion increment is 64 pixels. When a single expansion exceeds about 50% of the original image's width or height, the model has less visual context to work from and increasingly relies on guesswork — producing a generated area that diverges from the original's style. Expanding gradually (128–256 pixels per step) and then using "Use as Source" to continue from the result is significantly more stable than trying to extend 1024 pixels in a single operation. The maximum canvas size is 2048 pixels; requests beyond that are rejected.
What the edge blur parameter does
Edge blur (0–32) controls the width of the transition zone between the original image and the generated area. Higher values create a softer, less visible seam; values too low leave a visible line between the two regions. A value of 8–16 works for most natural scenes. If the original image has strong geometric lines — building edges, for instance — increase to 16–24 to mask the discontinuity.
Whether to fill in the prompt
Leaving the prompt empty lets the model extend entirely based on the image's existing colors, textures, and content. For large solid-color backgrounds, gradient skies, grass, or water, an empty prompt is usually sufficient.
Add a prompt when:
- The expansion direction contains little visual context for the model to reference
- You want to introduce specific elements in the extended area ("distant mountains," "city skyline," "cloud layer")
- You need to establish a specific mood or lighting quality
Avoid writing a prompt that contradicts the original image's content — expanding a daylight scene with a "night scene" prompt will produce an obvious tonal break at the boundary.
How iterative expansion works
After each expansion, the result can be loaded directly as the new source image using "Use as Source." The model then generates the next expansion using the already-extended content as context, which produces better continuity than starting from the original each time. Each expansion step is billed separately.