Turn a tight crop into a wider scene without reshooting anything. Upload a photo, pick the ratio or canvas size you actually need, and let an AI model paint in the missing edges so they match the color, texture and light of the original.
A tight crop is easy to fix after the fact, but a tight canvas rarely is. Portrait mode leaves no sky above your subject, a product photo arrives square when the banner needs 16:9, and an old family photo cuts your cousin's arm out of the frame. AI outpainting fills in what the camera never captured, matching the color, texture and light of the original so the new edge looks like it was always part of the scene.
Cropping only ever removes information; it cannot add a wider sky, more floor space or a second side of a room the camera never pointed at. For years the only fix was going back and reshooting, which is not an option for an old photo, a licensed stock image, or a product shot where the studio setup is long gone. Outpainting works the other way: it treats the edge of your photo as a starting point and generates new, plausible pixels beyond it, in the same style, color and lighting as what is already there.
Picasso IA runs this through models built for the job. Expand Image from Bria takes a photo and a target aspect ratio or exact canvas size and fills the new border with content that matches the original, with an optional prompt to steer what appears there. Black Forest Labs' Flux Fill Pro and Flux Fill Dev do the same job from another angle, built around masking and directional outpaint controls that also handle removing or replacing something inside the frame.
An outpainting model does not stretch or blur the pixels you already have, which is what a basic resize does. It reads the existing image as context, the same way a text model reads the start of a sentence, and generates new pixels consistent with everything visible near the edge: the direction of a shadow, the texture of a wall, the gradient of a sky.
Expand Image treats this as a canvas problem. Give it a source photo and either a preset ratio, such as 16:9 or 9:16, or an exact pixel size up to 5000 by 5000, and it places your original image inside that canvas and fills the surrounding space. A prompt field lets you describe what the new area should contain, useful for a specific addition rather than a plain continuation, and a negative prompt keeps unwanted elements out.
Flux Fill Pro and Flux Fill Dev take a mask-based approach. Paint a mask over blank canvas beside your image and describe what belongs there, or use the built-in outpaint presets, a left, right, top or bottom extension, or a 1.5x or 2x zoom out that pushes every edge back at once. Because these two also inpaint, the same tool can extend the canvas and then clean up something inside the original frame in a second pass.
The target format usually decides more of the outcome than the prompt does. Matching the ratio to where the image will run avoids a second round of cropping after expansion.
| Format | Typical use | What to type in the prompt |
|---|---|---|
| 1:1 square | Marketplace listing, profile grid | Leave blank for a plain continuation |
| 16:9 landscape | Website banner, desktop wallpaper | Describe what should fill the sides, such as more sky or floor |
| 9:16 vertical | Story, reel, phone wallpaper | Ask for headroom above or ground below the subject |
| 4:5 portrait | Social feed post | A short continuation cue, like matching wall or foliage |
| 3:2 print | Photo print, framed print | Leave blank unless the print crops close to an edge |
A blank prompt is often the right call: against a simple background, the model reads the existing texture and continues it convincingly on its own. Reach for the prompt field when the new area needs something specific, and use the negative prompt to rule out anything that keeps showing up uninvited.
This workflow covers the common case, reformatting one photo for a new ratio, and takes a few minutes from upload to download.
Pro tip: a huge jump, a tight square portrait pushed straight to a wide cinematic banner, asks the model to invent most of the final image at once. Two smaller expansions usually blend into the original better than one large one.
Reformatting is the most common case. A product photo shot square for a marketplace listing needs to become a 16:9 hero banner, or a print portrait needs a vertical version for a phone wallpaper, and outpainting delivers both without a second photoshoot. Real estate and interior photography benefit the same way: a slightly too-tight room shot can gain floor or ceiling to fit a listing template's fixed ratio.
Restoring old photographs is a second common use. Scans of prints trimmed decades ago, or family photos that were always a little too tight, can regain believable surroundings, though this stays a best-effort reconstruction rather than a recovery of what the camera actually saw. Print work is the third: enlarging the canvas around an image to add bleed and margin for framing is a smaller, more predictable version of the same technique, related to enlarging artwork for a bigger print size.
Product and marketing teams use it to keep one photoshoot working across every placement a campaign needs, square for the feed, vertical for a story, wide for a banner. If the background also needs cleaning up once it is extended, pairing this with background removal covers most of what a product image needs before it ships. Firefly's generative expand does a similar job inside Adobe's ecosystem; the comparison page covers how the two differ for a team weighing which to standardize on.
An honest look at the tool includes where it struggles, so you know when to expect a second pass or a manual touch-up.
None of this makes the tool unreliable for its actual job. It means the safest results come from modest expansions and a quick check of the seam before you call an image finished.
Outpainting is the generation of new content beyond a photo's original edges, so the picture covers a larger canvas or a different aspect ratio than it was shot in. Unlike stretching or cropping, which only reuse or discard pixels that already exist, an outpainting model generates new pixels that match the color, texture and lighting of the original scene, so the extended area reads as though the camera had simply captured more of it.
New Picasso IA accounts get free credits, and a single image expansion is one of the cheaper things to spend them on compared with video or 3D generation, so you can test the workflow on a real photo first. Beyond the free allowance, plans and current pricing are listed on the pricing page, since costs vary by model and change often enough that a specific number here would go stale.
Start with Expand Image for a straightforward result: give it a photo and a target ratio or exact canvas size, and it fills the new area automatically, with an optional prompt to steer the content. Reach for Flux Fill Pro or Flux Fill Dev when you need directional control, extending just one side, zooming out by a fixed multiple, or combining the expansion with an inpainting pass to fix something inside the original frame.
Yes, with a prompt field on both Expand Image and the Flux Fill models. Describing what the expanded area should contain, more sky, a matching floor, a specific background element, steers the generation toward that result, and a negative prompt excludes things you would rather not see. Leaving the prompt blank works well too, especially against a simple background, where the model reads the existing texture and continues it without extra instruction.
In most cases, yes, closely enough that the seam is not obvious at normal size. The model reads the color, texture and light near the edge of your source image and generates new pixels consistent with them, which is the entire point of the technique. Under bright, even lighting it matches especially well; under mixed or dramatic lighting, a faint shift can sometimes be visible on close inspection, which is why checking the seam at full size before exporting is worth the extra minute.
Expand Image accepts a custom canvas up to 5000 by 5000 pixels, well beyond what most web or print uses need, though the further you push beyond the original frame, the more of the final image is generated rather than captured, and results can look less anchored to the source. A large jump tends to work better split into two smaller expansions than attempted in one pass, since each step then has more original detail to work from.
Yes, Expand Image supports preserving the alpha channel, so a logo or a product cutout with a transparent background stays transparent through the expansion rather than gaining an opaque fill. This matters for design assets meant to sit on top of other backgrounds, since regenerating a solid fill where transparency should stay would defeat the purpose of the file.
Yes, and it is one of the more common uses in practice. A product shot photographed square for a marketplace listing can be expanded into a 16:9 banner or a 9:16 story format without a second shoot, and a logo placed on a small canvas can gain matching background space to fit a larger layout. The alpha-preserving option matters here more than for casual photos, since design assets frequently need to stay transparent.
Yes. Flux Fill Pro and Flux Fill Dev handle both inpainting, replacing or removing an area inside the existing frame, and outpainting, extending beyond it, so a photo can go through a mask-based cleanup and a canvas expansion without switching tools. Expand Image focuses on outpainting only, so for a photo needing an object removed inside the original frame, the Flux Fill models are the better start.
Bria trains Expand Image on fully licensed imagery so the output is safe for commercial projects, which matters for anyone reformatting product photography or marketing assets. As with any generated content, review the result before it ships, since an outpainting model occasionally produces a detail worth a second look, and check the current terms on the pricing page for anything specific to your plan.
A photo that is nearly the right shape is not the same as one that fits. Open the Picasso IA toolkit, upload the image cropped too tight for too long, and give it the edges it was always missing.
