How to Uncrop an Image with AI

Quick answer

You uncrop an image by uploading it, choosing a wider or taller target ratio, and letting an AI model paint the new area at the edges. The original stays as it is — nothing is cut. On Shotari's uncrop tool the built-in Basic model does this for free (2 a day without an account), and every paid model charges one extra credit when the target is 16:9 or 9:16.

How AI uncropping actually works

Most tools send your photo with an instruction like “expand this image.” That phrasing invites the model to re-render everything, and you get back a picture where the centre has subtly drifted — a face slightly different, a logo re-drawn wrong.

The gray-padding trick

Shotari does it differently, in the browser, before anything is sent. Your image is drawn onto a flat gray canvas (#c8c8c8) already sized to the target ratio, so the photo sits centred inside the empty margin it needs to fill. The prompt then asks the model to outpaint only the gray padding and to keep the central photo region “completely unchanged and pixel-identical.”

It is a small reframing with a big effect: the model is doing fill-in-the-blank rather than redraw-the-whole-thing, and the part you care about has a hard boundary it is told not to cross. If you are building this yourself with any image model, this is the part worth copying.

It only ever adds, never trims

The expansion direction is computed from your image, not chosen by you. If the target ratio is wider than what you uploaded, the canvas grows sideways and the full height is kept. If it is narrower, the canvas grows vertically and the full width is kept. Either way the original survives intact — going from a 4:3 photo to 16:9 adds strips on the left and right rather than slicing off the top and bottom the way a crop would.

What it costs per ratio

This is the part no other write-up can tell you, because it comes from how the pricing is actually computed. Uncrop always requests standard (1K) output, so resolution never multiplies the price — but the aspect ratio does. 16:9 and 9:16 carry a 1.1× multiplier, and the result is rounded up, which means one extra credit on every paid model. Square, 4:3, and 3:4 carry no multiplier at all.

Shotari's own two models are the exception — they are billed on a separate path that never applies the ratio multiplier, so Basic stays free and Pro stays 2 credits no matter how wide you go. Every other model on the list is a third-party model, and third-party generations can only be paid for with purchased credits: the 20 credits you get at signup and the 10 you can claim daily are free-tier credits, and they run Shotari Basic and Pro only.

Which ratio to target

Pick the ratio by where the image is going, then let the tool work out the pixels. These are the target sizes it offers for each shape:

A portrait photo taken at 3:4 that needs to become a 16:9 banner is asking the model to invent roughly twice the original width. That is a lot of guessing. Two smaller passes — 3:4 to 4:3, then 4:3 to 16:9 — give it less to invent each time and usually hold together better than one extreme jump. If the destination is a YouTube thumbnail, 16:9 is the only ratio worth targeting.

Prompts that work

The tool writes the outpainting instruction for you, so you do not have to. When you want more control — say the extended area should contain something specific — describe the surroundings, not the subject. The subject is already in the frame and does not need re-describing. The same principle runs through product photo work: constrain what changes, not what stays.

Where it still fails

Outpainting is confident even when it is wrong, so check these before you use the result anywhere that matters.

Re-running is the cheapest fix available. These models return a different result each time from the same input, so a scene that smears on the first attempt often lands cleanly on the third — which is another reason to try the free Basic model before spending credits on a premium one.