{"course":{"slug":"ai-images","title":"Image Generation, in Practice","url":"https://addaly.com/learn/ai-images","level":"beginner","track":"create"},"lesson":{"slug":"outpainting-the-crop","title":"Extending a frame when the crop changes","url":"https://addaly.com/learn/ai-images/outpainting-the-crop","position":46,"of":69,"minutes":8,"answer":"Outpainting extends an image in overlapping steps of a few hundred pixels, and the overlap is what lets the model match light, perspective and texture rather than inventing a discontinuous new scene.","keyIdea":"Outpainting extends an image in overlapping steps of a few hundred pixels, and the overlap is what lets the model match light, perspective and texture rather than inventing a discontinuous new scene."},"body":{"format":"markdown","text":"## The call that makes you need this\n\nThe image is approved at 4:5. Marketing now needs a 16:9 version for the website banner. Cropping a 4:5 frame to 16:9 removes most of the picture, and the subject goes with it.\n\nOutpainting adds canvas and fills it with content that continues the image. Done properly it is invisible. Done in one big jump it produces a seam, a change of light, and a background that does not agree with itself.\n\n## The method\n\n**Extend in steps of 128 to 256 pixels, not in one leap.**\n\n1. Enlarge the canvas by 256 pixels on the side you need.\n2. Mask the new empty area **plus 64 to 128 pixels of the existing image**: the overlap is the important part.\n3. Inpaint with denoise 0.8 to 1.0 over the new area, using a prompt describing what continues there, not the original scene prompt.\n4. Repeat until you reach the size you need.\n\nWhy steps: the model sees only its context window. A 512-pixel extension in one pass means most of the new region has no nearby real pixels to match against, so it invents freely: different light, different perspective, sometimes a different room. A 256-pixel step keeps every new pixel close to something real.\n\nWhy the overlap: without it, the model is generating a strip that must meet the original exactly at a hard boundary, and it will not. With 100 pixels of overlap it is *continuing* an image rather than abutting one, and the transition happens inside the regenerated area where nobody can see it.\n\n## Prompting an extension\n\nDescribe what should be there, not what the image is of.\n\n- Extending upward from a courtyard scene: `whitewashed wall continuing upward, overcast sky above the roofline`, not the full original prompt.\n- Extending sideways from a portrait: `plain studio background, soft falloff`, not the description of the person, which will get you a second person.\n\nThat second-person failure is the most common outpainting mistake. The prompt still describes a woman, the model has empty canvas, and so it puts a woman there.\n\n## What breaks\n\n**Perspective drift.** Each step can shift the vanishing point slightly, and after four steps a straight wall curves. Check long straight lines after every step rather than at the end. If drift starts, a control (canny on the existing edges, or a depth map) holds the geometry.\n\n**Light drift.** The new area gradually becomes brighter or flatter. Same cause, same remedy: check each step, and correct the grade as you go rather than at the end.\n\n**Repetition.** Extending a patterned surface produces obvious repeats, because the model continues what it sees. Vary the prompt slightly on each step, or break it up with a repair pass afterwards.\n\n**The tell at the join.** Look specifically for a faint vertical or horizontal line where the seam was, visible as a slight change in noise or contrast. A low-denoise pass at 0.2 over a band spanning the join clears it.\n\n## Doing it in one operation\n\nSome tools handle this well in a single step: Photoshop's generative expand, InvokeAI's canvas, and several hosted services all extend a frame cleanly in one action. If you have one and the extension is modest, use it.\n\nThe manual method still matters for two reasons. Large extensions (doubling a frame's width) fail in one shot regardless of tool, because the problem is context rather than implementation. And the manual method gives you control over the prompt at each step, which is how you decide what is in the new space rather than accepting what the model thinks belongs there.\n\n## The better answer, when there is time\n\nOutpainting is a rescue. The better outcome is not needing it, and that is the brief block's job: ask at the start which ratios are required, and either generate wider than any of them or generate each ratio separately.\n\nThere is also a real ceiling. Outpainting extends a *scene*; it cannot extend a *composition*. If the original frame worked because the subject was placed against the edge, adding 400 pixels of room changes that relationship, and the wider version can be technically seamless and compositionally worse. Look at the extended image as a new picture and judge it as one, rather than only checking whether the join shows. Sometimes the honest answer to the banner request is a fresh generation at 16:9 using the approved frame as a reference, and it takes less time than four careful extension steps."},"check":{"question":"Why does outpainting in 256-pixel steps with overlap work better than one 512-pixel extension?","options":[{"text":"Because a smaller masked area allows a higher denoise value to be used safely, which produces more convincing new content in the extended region.","correct":false,"tempting":"Denoise is already near 1.0 in both cases, since the area is empty; step size is not what permits a high value."},{"text":"Because each new pixel stays close to real pixels it can match light and perspective against, and the overlap hides the join inside regenerated area.","correct":true},{"text":"Because the model processes the image in fixed tiles, so an extension larger than one tile is composited from separate passes that cannot agree with each other.","correct":false,"tempting":"Tiling is a different mechanism used for large canvases; ordinary outpainting is a single pass over the enlarged frame."},{"text":"Because repeated smaller passes allow the prompt to be adjusted between steps, which is the main reason the incremental approach produces better results.","correct":false,"tempting":"A genuine benefit of the method and a secondary one; the same prompt used throughout still gives a far better result in steps than in one leap."}]},"prev":{"slug":"text-in-images","title":"Type is set, not generated","url":"https://addaly.com/learn/ai-images/text-in-images"},"next":{"slug":"removing-a-thing-properly","title":"Four ways to remove an object, and when each is right","url":"https://addaly.com/learn/ai-images/removing-a-thing-properly"},"authorship":"Founder and series editor: Dr Mustafa Mun. Written by AI models to a written standard. Exam keys checked blind by a second model: 3,665 of 3,666 agreed. Academic review invited.","review":{"status":"Draft, open for academic review"},"cite":"Addaly (2026) 'Extending a frame when the crop changes', in Image Generation, in Practice. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-images/outpainting-the-crop (Accessed: 13 October 2026).","licence":{"name":"CC BY-NC-SA 4.0","url":"https://creativecommons.org/licenses/by-nc-sa/4.0/"},"reader":{"schema":1,"title":"Extending a frame when the crop changes","courseTitle":"Image Generation, in Practice","module":{"index":5,"slug":"repair-and-extend","title":"Repair, extend, assemble"},"answer":null,"keyIdea":"Outpainting extends an image in overlapping steps of a few hundred pixels, and the overlap is what lets the model match light, perspective and texture rather than inventing a discontinuous new scene.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"the-call-that-makes-you-need-this","runs":[{"text":"The call that makes you need this"}]},{"t":"p","runs":[{"text":"The image is approved at 4:5. Marketing now needs a 16:9 version for the website banner. Cropping a 4:5 frame to 16:9 removes most of the picture, and the subject goes with it."}]},{"t":"p","runs":[{"text":"Outpainting adds canvas and fills it with content that continues the image. Done properly it is invisible. Done in one big jump it produces a seam, a change of light, and a background that does not agree with itself."}]},{"t":"h2","id":"the-method","runs":[{"text":"The method"}]},{"t":"p","runs":[{"text":"Extend in steps of 128 to 256 pixels, not in one leap.","b":true}]},{"t":"ol","start":1,"items":[[{"text":"Enlarge the canvas by 256 pixels on the side you need."}],[{"text":"Mask the new empty area "},{"text":"plus 64 to 128 pixels of the existing image","b":true},{"text":": the overlap is the important part."}],[{"text":"Inpaint with denoise 0.8 to 1.0 over the new area, using a prompt describing what continues there, not the original scene prompt."}],[{"text":"Repeat until you reach the size you need."}]]},{"t":"p","runs":[{"text":"Why steps: the model sees only its context window. A 512-pixel extension in one pass means most of the new region has no nearby real pixels to match against, so it invents freely: different light, different perspective, sometimes a different room. A 256-pixel step keeps every new pixel close to something real."}]},{"t":"p","runs":[{"text":"Why the overlap: without it, the model is generating a strip that must meet the original exactly at a hard boundary, and it will not. With 100 pixels of overlap it is "},{"text":"continuing","i":true},{"text":" an image rather than abutting one, and the transition happens inside the regenerated area where nobody can see it."}]},{"t":"h2","id":"prompting-an-extension","runs":[{"text":"Prompting an extension"}]},{"t":"p","runs":[{"text":"Describe what should be there, not what the image is of."}]},{"t":"ul","items":[[{"text":"Extending upward from a courtyard scene: "},{"text":"whitewashed wall continuing upward, overcast sky above the roofline","code":true},{"text":", not the full original prompt."}],[{"text":"Extending sideways from a portrait: "},{"text":"plain studio background, soft falloff","code":true},{"text":", not the description of the person, which will get you a second person."}]]},{"t":"p","runs":[{"text":"That second-person failure is the most common outpainting mistake. The prompt still describes a woman, the model has empty canvas, and so it puts a woman there."}]},{"t":"h2","id":"what-breaks","runs":[{"text":"What breaks"}]},{"t":"p","runs":[{"text":"Perspective drift.","b":true},{"text":" Each step can shift the vanishing point slightly, and after four steps a straight wall curves. Check long straight lines after every step rather than at the end. If drift starts, a control (canny on the existing edges, or a depth map) holds the geometry."}]},{"t":"p","runs":[{"text":"Light drift.","b":true},{"text":" The new area gradually becomes brighter or flatter. Same cause, same remedy: check each step, and correct the grade as you go rather than at the end."}]},{"t":"p","runs":[{"text":"Repetition.","b":true},{"text":" Extending a patterned surface produces obvious repeats, because the model continues what it sees. Vary the prompt slightly on each step, or break it up with a repair pass afterwards."}]},{"t":"p","runs":[{"text":"The tell at the join.","b":true},{"text":" Look specifically for a faint vertical or horizontal line where the seam was, visible as a slight change in noise or contrast. A low-denoise pass at 0.2 over a band spanning the join clears it."}]},{"t":"h2","id":"doing-it-in-one-operation","runs":[{"text":"Doing it in one operation"}]},{"t":"p","runs":[{"text":"Some tools handle this well in a single step: Photoshop’s generative expand, InvokeAI’s canvas, and several hosted services all extend a frame cleanly in one action. If you have one and the extension is modest, use it."}]},{"t":"p","runs":[{"text":"The manual method still matters for two reasons. Large extensions (doubling a frame’s width) fail in one shot regardless of tool, because the problem is context rather than implementation. And the manual method gives you control over the prompt at each step, which is how you decide what is in the new space rather than accepting what the model thinks belongs there."}]},{"t":"h2","id":"the-better-answer-when-there-is-time","runs":[{"text":"The better answer, when there is time"}]},{"t":"p","runs":[{"text":"Outpainting is a rescue. The better outcome is not needing it, and that is the brief block’s job: ask at the start which ratios are required, and either generate wider than any of them or generate each ratio separately."}]},{"t":"p","runs":[{"text":"There is also a real ceiling. Outpainting extends a "},{"text":"scene","i":true},{"text":"; it cannot extend a "},{"text":"composition","i":true},{"text":". If the original frame worked because the subject was placed against the edge, adding 400 pixels of room changes that relationship, and the wider version can be technically seamless and compositionally worse. Look at the extended image as a new picture and judge it as one, rather than only checking whether the join shows. Sometimes the honest answer to the banner request is a fresh generation at 16:9 using the approved frame as a reference, and it takes less time than four careful extension steps."}]}],"check":{"question":[{"text":"Why does outpainting in 256-pixel steps with overlap work better than one 512-pixel extension?"}],"options":[{"text":[{"text":"Because a smaller masked area allows a higher denoise value to be used safely, which produces more convincing new content in the extended region."}],"correct":false,"tempting":[{"text":"Denoise is already near 1.0 in both cases, since the area is empty; step size is not what permits a high value."}]},{"text":[{"text":"Because each new pixel stays close to real pixels it can match light and perspective against, and the overlap hides the join inside regenerated area."}],"correct":true,"tempting":null},{"text":[{"text":"Because the model processes the image in fixed tiles, so an extension larger than one tile is composited from separate passes that cannot agree with each other."}],"correct":false,"tempting":[{"text":"Tiling is a different mechanism used for large canvases; ordinary outpainting is a single pass over the enlarged frame."}]},{"text":[{"text":"Because repeated smaller passes allow the prompt to be adjusted between steps, which is the main reason the incremental approach produces better results."}],"correct":false,"tempting":[{"text":"A genuine benefit of the method and a secondary one; the same prompt used throughout still gives a far better result in steps than in one leap."}]}],"order":[2,3,1,0],"why":null},"prev":{"slug":"text-in-images","title":"Type is set, not generated","minutes":8},"next":{"slug":"removing-a-thing-properly","title":"Four ways to remove an object, and when each is right","minutes":8}}}