Outpainting a scene whose extension needs real structure, such as the rest of a building, and accepting whatever the model invents.
AI and on-device processing
Generative fill
Also called ai fill, outpainting or generative expand. Here is what it means, when it changes what you export, and what people get wrong.
AI and on-device processingRuns on your deviceUpdated 10 August 2026
The short answer
Generative fill uses a text prompt and a masked area to have a model invent new image content, either inside the frame or beyond its edges. Outpainting is the same idea applied outwards, extending a photo so it fits a different aspect ratio without cropping the subject.
Updated
The short version
Generative fill at a glance
| Quick fact | Detail |
|---|---|
| Inputs | Image, mask, and usually a text prompt |
| Outpainting | Generates content beyond the original frame |
| Common use | Extending a backdrop to a new aspect ratio |
| What it produces | Plausible invented detail, not recovered detail |
| Works best on | Plain backdrops, sky, blurred surfaces |
| Marketplace limit | Main listing images must show the real product |
Slide the table sideways to see every column.
How is it different from ordinary inpainting?
Plain inpainting has one goal: make the hole disappear. It is judged on whether the repair is invisible, and it takes all its cues from the pixels around the mask.
Generative fill adds a second input, your prompt, and a different goal: put something specific there. The model is still conditioned on the surrounding image, so lighting and perspective roughly carry over, but it is now free to introduce objects, surfaces and detail that appear nowhere in the original. That freedom is why it can extend a scene convincingly and also why it will confidently produce a chair with five legs.
When is outpainting the right way to change aspect ratio?
You often have a photo in the wrong shape: a portrait shot that a marketplace wants square, or a square product image you need at a taller ratio for a social placement. The usual options all cost something. Cropping loses part of the subject. Letterboxing wastes space and can trip layout rules. Stretching distorts everything.
Outpainting adds new canvas and fills it with plausible continuation of the existing background, keeping the subject intact at its original framing. It works best when the added region is simple: seamless backdrop, a wall, sky, an out-of-focus surface. It works badly when the extension has to invent structure, such as the rest of a room, the rest of a face, or the missing half of a logo.
Is it safe to use generated content in a product photo?
Be careful here, because the rules are stricter than the technology. Marketplace image policies are built around the main image showing the actual item. Amazon's product image requirements state that the main image must be on a pure white background, with the product filling most of the frame and no added text, logos, watermarks or borders. eBay's picture guidance likewise rules out added graphics, badges, logos and watermarks.
Practically, that gives you a clear line. Extending a plain backdrop so the subject sits correctly in a square frame is a framing operation and is generally fine. Generating a lifestyle scene around the product, inventing packaging, adding accessories that are not in the box, or letting the model reshape the product itself is a misrepresentation of what the buyer receives, and it is the fastest route to returns and a policy strike.
What people get wrong about generative fill
Each one is a real failure mode, not a style preference.
Using a generated lifestyle background on a main listing image where the platform requires the product on plain white.
Prompting for the whole picture rather than for the masked region, which is the only part the model is being asked to produce.
Forgetting that the fill inherits nothing of the original grain or lens character, so an extended edge looks cleaner than the photo it joins.
Questions people ask
Generative fill, answered
The follow-up questions people search for once they have the definition.
Does generative fill change the parts of the photo I did not mask?
It should not, but many implementations re-encode the whole image on export, so unmasked pixels can pick up a second round of compression. If fidelity matters, check the output at full zoom against the original, and prefer tools that composite the fill back into the untouched source.
Why does the extended area look sharper than my photo?
Because it was generated rather than captured. It has no lens softness, no sensor noise and no compression history, so it can look suspiciously clean next to real pixels. Adding a small amount of matching grain and a slight blur across the seam usually settles it.
Can I use outpainting instead of reshooting for a square listing image?
For a plain studio backdrop, usually yes, and it is far quicker than a reshoot. If the extension would need to invent part of the product, the surface it sits on, or a recognisable environment, reshoot or pad with a flat colour instead.
Do I have to disclose that an image was edited this way?
Rules vary by platform and by what was changed. Adding empty backdrop is treated very differently from altering the product itself. Check the specific marketplace or advertising policy that applies to you, and assume anything that changes what the buyer sees they are getting needs to be accurate.
Same cluster
More from the glossary
Neighbouring entries, so every term in the set is one hop from every other.
Geotag
A geotag is the location a camera or phone wrote into a photo's EXIF metadata: latitude, longitude, usually altitude and a GPS timestamp.
Gamma
Gamma is the non-linear relationship between the numbers stored in an image and the amount of light a display emits.
Ghost mannequin
Ghost mannequin is an apparel photography technique where a garment is shot on a mannequin, then the mannequin is edited out so the clothing appears to hold its own shape.
From the studio
Tools that do this
Each one runs on your own device, so the photo stays where it is.
Sources
Where the facts on this page come from. Every link opens the specification, standard or documentation the claim was read at.
Keep reading
Elsewhere on Photomop
Platform image rules
Image size, file type and background rules for 22 selling platforms, each one sourced.
Read itConvert image
JPG, PNG, WebP and AVIF, with an honest answer about HEIC and alpha.
Read itAll the tools
The whole toolbox in three families, with what each one needs before it can run.
Read itOn-device privacy
The pipeline in five steps, what leaves your device, and how to check it yourself.
Read it
Now do it to a photo
Photomop is a photo studio that runs on your own device. Resize, crop, compress, convert, batch edit, remove a background and change one all work in the browser tab you are reading this in, at full resolution and with no watermark. The on-device page shows you how to check that the photo stays put.