AI Image Expand
Add canvas around a photo instead of cropping it. The model paints the missing edges so the picture fits a wider frame, banner or print without stretching the subject.
- Nothing stretched
- Subject kept intact
- JPG, PNG, WEBP
- No watermark
Free credits to start. No watermark.
Image expand solves a problem cropping cannot: the frame is wrong, but the subject is not. A vertical phone photo needs to become a 16:9 thumbnail, a product shot needs empty space on the side for a headline, a portrait needs air above the head. Cropping fixes the shape by throwing pixels away and usually cuts someone out of frame. Expansion keeps every original pixel, adds new canvas around it, and lets the model continue the background into the space that did not exist before.
The technique is called outpainting, and the honest way to judge it is by what it does not do: it does not stretch the subject, it does not redraw the face, and it does not move anything inside the original frame. Everything new happens outside the border, which is why the result still looks like the same photograph rather than a generated scene.
Image expansion vs cropping vs stretching
| Approach | What happens to the subject | What happens to the frame | When to use it |
|---|---|---|---|
| Crop | Pixels are removed, subject may be cut | Smaller canvas | You need the shape and can lose the edges |
| Stretch | Proportions distort, faces widen | Same canvas, wrong aspect | Never for people or products |
| Expand (outpaint) | Subject untouched, same scale | New canvas, background continues | Wrong shape, right subject, room needed for text |
If the subject already fills the frame edge to edge, expansion has nothing to continue from. Leave a little background on the sides you want to extend.
How to expand a picture in three steps
- 1Upload the photo you want to keepJPG, PNG or WEBP up to 4 MB. The tool reads your original dimensions and keeps them as the anchor for everything it adds.
- 2Choose the new shape1:1 for feed posts, 4:5 for portrait posts, 16:9 for banners and thumbnails, 21:9 for cinematic frames. The output size updates instantly so you know the target before you run it.
- 3Generate, then look at the edgesThe before and after panels sit side by side. Judge the join line: the new background should match the grain, the light direction and the horizon. If it does not, add a hint in the note and run it again.
What expand picture size works best on
- Skies, plain walls, floors, grass, water and studio backdrops, because the continuation is predictable.
- Product shots on a seamless background, where the added canvas is nearly invisible.
- Portraits that need headroom for a title or a logo lockup.
- Vertical phone video frames that need to become 16:9 for a thumbnail.
- Real estate and interiors where a wider frame sells the room better than a tighter crop.
- Old prints that are too tightly cropped, where the surrounding paper texture can be rebuilt.
When expansion gets hard
Two situations are genuinely difficult. The first is hard geometry: brick walls, tiled floors, railings, windows with mullions. The model must continue a repeating pattern with perspective, and small errors are obvious. The second is a busy edge where the subject touches the frame border, because there is no context for what should appear behind them.
Both cases improve with a note. "Continue the brick wall in the same perspective" gives the model the pattern it needs; "extend the beach and keep the horizon level" stops it inventing a second horizon. If a run still fails, expanding in two smaller steps (16:9 first, then 21:9) is often more reliable than one large jump.
Aspect ratios and where each one is used
| Shape | Typical use | Notes |
|---|---|---|
| 1:1 | Instagram feed, marketplace listing, profile post | The safest default; most platforms crop nothing. |
| 4:5 | Portrait feed post, print flyer | Takes more vertical space in feed, suits full-body shots. |
| 16:9 | YouTube thumbnail, website hero, presentation | The most requested shape for anything with a headline. |
| 21:9 | Cinematic still, banner, wide email header | Needs a background with a strong horizontal flow. |
Tips for a clean expansion
- Start from the sharpest original: expansion adds detail but cannot recover a soft source.
- Give the model a direction when the scene has one ("continue the road", "extend the sky").
- Avoid adding text inside the new area; compose the image first, add type afterwards in a design tool.
- If the subject sits off-centre, say so, and ask for the empty space on the side you actually need.
- For a product set, keep the same background and the same model so the extra canvas matches across images.
Free AI image expansion, credits and limits
Expansion runs on paid image models, so it needs an account: the price per model is shown before you run anything, and the cheapest option works on the free credits that come with a new account. Uploads are capped at 4 MB and the output is rendered at up to 1344 pixels on the longest edge in the shape you chose.
For a large final file, expand first and then use AI image upscale, which is free and takes the result to 4096 pixels without spending credits. If the source is an old print, restore it before expanding, or the new canvas will match the damage instead of the picture.
Outpainting versus generative fill: same idea, different control
Every tool that adds canvas is doing some version of outpainting, but the control you get varies enormously. Some let you paint a mask and generate a small patch; others take the whole frame and a target shape and continue everything at once. Masks are precise but slow, and they are the right choice when one specific corner needs work. Whole-frame expansion is faster and more consistent, because the model sees the full composition and continues the light, the horizon and the grain in one pass.
This page takes the whole-frame route, which is why the choices are aspect ratios rather than brush tools. For a thumbnail, a banner or a marketplace image, the question is almost always what shape the destination needs, not which five pixels to invent. If you do need local control, expand first and then retouch the join in an editor.
Fixing the join line when it shows
A visible seam is nearly always a mismatch in one of three properties: grain, light direction, or perspective. Grain is the easiest to spot on smooth areas such as sky or studio backdrops, where the original has noise and the new area is perfectly clean, or the reverse. Light direction shows where a shadow falls the wrong way. Perspective shows in any repeating structure, where the lines in the new area do not converge to the same vanishing point.
The fix is a hint, not a heavier prompt. Name the property: “match the grain of the original”, “keep the light coming from the left”, “continue the floor tiles in the same perspective”. If it still shows, generate again at a smaller ratio step, since the model has less empty space to fill and more context per new pixel.
| Problem | What it looks like | Prompt hint that helps |
|---|---|---|
| Grain mismatch | Smooth new area next to noisy original | Match the film grain and texture of the original |
| Light mismatch | Shadow falls the wrong way in the new area | Keep the same light direction and colour temperature |
| Perspective break | Repeating lines bend at the seam | Continue the wall and floor in the same perspective |
| Subject clipped | Head or foot touches the new border | Add breathing room, keep the subject fully inside |
Expansion for video and thumbnails
Vertical phone video is the most common mismatch on the internet: shot at 9:16, published at 16:9, and the platform either crops the face or adds black bars. Expanding a still frame to the target shape gives you a clean background you can reuse as a thumbnail plate, and the same idea works for podcast clips, where a square frame only needs the sides filled with a blurred or extended room.
For thumbnails specifically, expansion is a layout tool. Generate the 16:9 frame first, then place text and faces in the empty space you deliberately created, rather than generating text inside the image, which almost always looks wrong and cannot be edited afterwards.
Marketplace sizes and social safe areas
Each platform crops differently, and knowing the safe area saves a second run. Instagram feed tolerates a 4:5 portrait and crops nothing; a 16:9 image in the same slot gets letterboxed or centre-cropped. YouTube thumbnails are 16:9 at 1280 by 720, with the duration badge covering the bottom right, so the subject should sit left of centre. Website heroes are usually 16:9 or wider, and text overlays occupy the left third on desktop and the bottom half on mobile.
Expansion is the tool that makes a single photo fit all of those without reshooting. Generate the widest version you need, then crop down for the narrower slots, because cropping a correct image is lossless in composition terms while expanding into a wrong shape usually shows at the seam.