Image editing
What is outpainting?
Outpainting extends an image beyond its original borders. An AI model adds new content around the picture that continues its scene, light and perspective, so a square photo can become a wide banner or a tall phone wallpaper without stretching or cropping the original. It is also called image expansion or uncropping.
By the DreamBench team. Updated .
Outpainting at a glance
At a glance
- Input
- A picture and the new frame it should fill.
- Output
- A larger picture: the original in it, new scene around it.
- Keeps
- The original pixels, when the tool lays the original back over the result instead of redrawing it.
- Invents
- Everything outside the original edges.
- Struggles with
- People cut off at the edge, text, strong patterns, very large expansions.
Which do I need: expand, reframe or crop?
More room around the subject: Outpainting: the picture stays whole and new scene is generated around it. The AI image expander does it in the frame you pick, your original kept untouched in the middle.
A different shape for a post, a story or a video: Outpainting into that frame, so nothing is cut off: change the image aspect ratio picks the frame for the platform and extends, pads or crops the photo to fit it. 9:16 for a story or Reel, 16:9 for a video, 1:1 for a square post.
Only the part I already have, tighter: A crop, which loses the pixels outside the new frame and invents nothing. Use it when the subject fills the frame already, or when nothing invented may appear in the picture.
More detail at the same frame: Not outpainting but upscaling: more pixels for the same picture, the detail predicted. The AI image upscaler does it at 2x or 4x.
How outpainting works
Outpainting works by placing the original picture on a larger canvas with the empty area marked. The model reads the edges that touch that area, the horizon, the floor, the direction of light, and generates what would plausibly continue them: the sky goes on, the floor continues, a table gains its other end.
Good outpainting leaves the original pixels as they were and matches the new area to them in light, grain and perspective. Everything it adds is invented: what lay beyond the frame was never photographed, so the result is a plausible scene, not a recovered one.
The new area is where it breaks. The model has the old edge to read and nothing else, so a face, a hand or lettering cut by that edge is finished by guesswork, and a busy pattern expanded by a lot can repeat itself or soften toward the far edge. A photo with sky, floor or a wall at its edges expands convincingly; one with a person touching the border is the hard case, so frame the subject away from it first, and expand in smaller steps if the far edge goes soft.
Ways to change a picture’s shape
| The original | What is new | |
|---|---|---|
| Cropping | Loses whatever falls outside the new frame | Nothing |
| Stretching | Kept whole, but distorted | Nothing |
| Padding | Kept whole and undistorted | Plain bars of color or blur |
| Outpainting | Kept whole and undistorted | More of the scene, generated |
To fill a gap inside the picture instead, the job is inpainting. To add detail without changing the frame, it is upscaling.
How to get a good outpainting
- Pick the frame you need first: 16:9 for a video or a banner, 9:16 for a story.
- Check the edges of the original. A person or an object cut off at the edge has to be invented to complete it, which is where joins show.
- Prefer pictures with room around the subject: sky, floor, a wall, a landscape.
- Expand modestly. The more of the frame is new, the more the model invents.
- Compare the result with the original at full size, along the old edges, for light, grain and perspective.
Where people use outpainting
- Changing the aspect ratio: a 4:3 photo into a 16:9 video frame or a 9:16 story, without cropping the subject.
- Banners and headers that need more room on either side.
- Product shots that need space around them for text.
- A photo framed too tightly, given room to breathe.
How DreamBench does outpainting
The AI image expander places your photo in the middle of the frame you pick, and FLUX.2 [pro] draws only the new border around it. If that model declines a photo, Nano Banana 2 continues the scene instead. Either way your original is laid back over the middle of the result, so the part you uploaded stays as it was.
The price depends on the size of the photo and is shown on the button before you run it. The wider picture can open in the composer as the first frame of a video.
At a glance
- Frames
- 16:9, 9:16, 1:1, 4:3, 3:4, 4:5 and 21:9. All but 3:4 and 4:5 are frames DreamBench’s video models render.
- Input
- JPEG, PNG, WebP or HEIC up to 40 MB. The photo is scaled down so that it and the wider canvas fit the model’s working size, which is why a large upload can come back with fewer pixels.
- Models
- FLUX.2 [pro] draws the new border; Nano Banana 2 continues the scene when it declines a photo.
- Direction
- Centered: the photo sits in the middle and the new scene is added equally on the sides the frame needs.
- Prompt
- None. The scene is continued as it is: more of the same sky, room or street.
- Output
- A JPEG in the frame you pick, your original laid back over the middle, uncropped. No transparent background.
- Price
- Depends on the size of the photo and the frame, and is shown on the button before you run it; a run that fails is refunded.
- Next step
- Animate this frame opens the composer with the result as the reference for a clip.
Questions
What is the difference between outpainting and inpainting?
Inpainting fills a region inside a picture, such as the gap left by a removed object. Outpainting adds content outside the original edges to make the picture larger. Both generate new content that matches what is already there.
Does outpainting change the original photo?
It shouldn’t. Careful outpainting keeps the original area untouched and only adds around it. DreamBench’s expander lays your original back over the result, so the part you uploaded stays exactly as it was.
Can I use outpainting to change the aspect ratio of an image without cropping it?
Yes, outpainting changes a picture’s aspect ratio without cropping, and that is its most common use: a 4:3 photo becomes 16:9 for video or 9:16 for a story, with the subject untouched and new scene generated around it. Cropping loses pixels and stretching distorts; outpainting keeps the original whole and adds to it.
How do I extend a photo with AI?
Pick the frame you need, place the photo in it and let an outpainting model generate the border around it. On DreamBench the image expander does this: your original stays in the middle untouched and only the new area is generated.
Can I tell the expander what to add, or extend only one side?
No. DreamBench’s expander takes no prompt: it continues the scene from the edges, so pick a photo whose edges show what you want more of. It also centers your photo in the new frame, so a landscape going to 9:16 grows up and down, and a portrait going to 16:9 grows left and right.
What makes an outpainted edge look natural or obvious?
It looks natural when the new area matches the original’s light, grain and perspective. Joins show on large expansions, on busy patterns and where the original edge cuts through a face, a hand or text: a person cut off at the edge is invented to complete them, so check the result at full size along the old edge.
How much can I expand a picture?
Modestly, by preference: the more of the frame is new, the more the model invents. Expand in two steps rather than one large one if you need a lot. On DreamBench a large photo is scaled down first so the wider canvas fits, so the result can have fewer pixels than the upload; upscale it afterward for a large print.
Is outpainting useful for video production?
Yes. A picture expanded to 16:9 or 9:16 can be the first frame of an image-to-video clip. Outpainting a whole video, frame by frame, is a separate and harder job.
Is outpainting the same as uncropping?
Yes, in practice. Uncropping, image expansion and outpainting all describe the same job: extending a picture past its edges with new content that matches it.
Which aspect ratios can I expand a picture to?
On DreamBench, seven: 16:9, 9:16, 1:1, 4:3, 3:4, 4:5 and 21:9. All but 3:4 and 4:5 are frames DreamBench’s video models render, so a picture expanded to the frame a clip will use is a reference that already fits; 4:5 is for a portrait post. There is no custom canvas: the photo sits in the middle and the frame decides where the new scene goes.
What file does the expander return, and does it keep transparency?
A JPEG in the frame you pick, with your original laid back over the middle, uncropped. A JPEG has no alpha channel, so the result is never transparent: to keep a cutout see-through, place it on a larger transparent canvas in an editor instead, or remove the background after you expand. A large upload is scaled down first, so the result can have fewer pixels than the upload.
What happens if the expander declines my photo?
If FLUX.2 [pro] declines a photo under its content rules or a moderation check, DreamBench tries Nano Banana 2 once in its place: it continues the scene on a gray canvas of the frame you picked, and your original is laid back over the middle in the same way. The run keeps the price shown on the button, and it is refunded only when both models decline. A plain failure is retried rather than passed on.
Does outpainting cost anything?
Some editors include a basic expand tool. On DreamBench you create a free account and choose a plan; the price of an expansion depends on the photo’s size and is shown on the button before you run it.
Related
DreamBench tools
Terms
- InpaintingFilling a chosen area of a picture with new content, the way AI removes objects and text.
- Image upscalingMaking a picture larger with more pixels, the extra detail predicted so it stays sharp.
- Image to videoAI video that starts from a still picture, the motion generated from a prompt.
- Transparent backgroundPixels with no color of their own, stored in an alpha channel, so the page behind shows through.
Sources
This page describes DreamBench’s own tools and models, from the same tables the product uses.