Image editing from a prompt

Upload a photo and describe the edit in plain words. The AI changes exactly what you asked for: the light, textures, faces and geometry of the frame stay put. No masks, no layers, no selections.

Edit a photo from a description

Recolor the sofa deep blue, keep the fabric and the folds

One phrase — and the frame is different

On the left is what went into the AI, on the right is what came back. Under each pair is the phrase the edit was asked with. The source frames are generated too: we don't want to show off the editor on other people's photographs.

Replace the background with a clean studio scene: a light grey gradient and a soft shadow under the sneakers. Keep the sneakers exactly as they are — same white color, same shape, same texture

Remove all the people from the embankment, keep only the landscape

Recolor the sofa deep blue, keep the fabric and the folds

Turn the selfie into a neat business portrait: soft light, blurred background, natural retouching

Restore the old photograph: remove the scratches, stains and the tear, bring back the sharpness

Colorize the black-and-white photo in natural colors, keep the film grain

Furnish the room: a light sofa, a round coffee table, a floor lamp, a rug and a big plant

Make it an evening scene: sunset sky, warm light in the windows and street lamps, wet asphalt

Turn summer into golden fall: yellow and orange foliage, fallen leaves on the path

Remove the wires, the rubbish bin and the traffic cone, keep the facade and the sky

Replace the sky with a dramatic sunset with clouds, light the facade with warm light

Turn the sketch into a photorealistic render of the kitchen: oak, white stone, daylight

Print the name ROASTLAB on the bag, below it the line “ethiopia · single origin” and a small coffee bean icon

Dress him in a navy business suit with a white shirt, don't change the face or the pose

Set the bottle on a surface, add a soft shadow and a slight reflection, make the background warm beige

Turn the photograph into a watercolor illustration with soft washes and white paper

Make it a holiday shop window: a garland, fir branches, snow on the windowsill, warm light

Try it on your own photo

What the editor can do

One tool instead of a dozen: the things people used to open a graphics editor and call a retoucher for.

Edits precisely instead of redrawing

The AI changes only what you asked for. The product stays the same product, the face the same face, and the perspective and light adjust to the edit on their own.

Understands plain English

No masks, brushes or selections. “Remove the wires”, “recolor it blue”, “put it on a beige background” — that's enough to get the result.

Color, light and materials

Change the product's color, the time of day, the season or the weather. Shadows and reflections are recalculated with the scene, so the frame doesn't fall apart.

Works with text in the picture

A caption on a package, a sign or a price tag can be added or replaced — the letters come out even and readable.

Edits stack one after another

The result can be edited further with the same kind of phrase: first remove the clutter, then swap the background, then add the text. The history of steps stays in the project.

How to edit a photo from a description

Four steps from the source shot to the finished frame — all in the browser, nothing to install.

1
Sign up for SUPA and create a project or open an existing one
2
Upload the photo you want to fix and select it on the canvas
3
Open the AI section and pick prompt-based image editing
4
Describe the edit in a phrase or two and run the generation. Accept the result, regenerate it, or edit it further

How to phrase an edit to get it right the first time

Say what to change — and what to leave alone

The model happily improves everything in sight unless you stop it. The most useful half of the prompt is what must stay the same.

Bad: “make it pretty”
Good: “replace the background with a light grey gradient; keep the product itself — color, shape and texture — unchanged”

One edit — one request

Five demands in one phrase almost always produce a compromise where none is carried through. Remove the clutter, then swap the background, then add the caption — three short steps work better than one long one.

Name objects the way they look in the frame

“Remove the rubbish bin and the orange cone on the right” is more reliable than “remove the junk”. When there are several objects, a location helps: on the left, in the foreground, by the wall.

Describe the light, not just the object

A frame looks real when the edit matches the lighting. “A soft shadow under the object”, “warm light from the window”, “an overcast day” — short clarifications you see on the result immediately.

For text, give the exact wording

Write the line in quotes exactly as it should appear and say where to put it: “ROASTLAB large on the bag, below it a small line ‘ethiopia · single origin’”. The shorter the line, the cleaner the result.

If the result is off, change the wording, not the number of attempts

The same phrase gives similar results. Add the detail that was missing: the color, the material, the angle, the boundary of the edit — that's usually enough.

Frequently asked questions about AI photo editing

What photos can I upload?

Any raster images — JPG, PNG, WebP. The larger and sharper the source, the neater the edit: don't expect miracles from a 300 × 300 shot.

Will the rest of the photo change?

The AI redraws the frame as a whole but tries to preserve everything it wasn't asked about. In practice small background details may shift a little, so for important shots compare the result with the source — both versions stay in the project.

Can I edit people?

Yes: change the outfit, fix the light, turn a selfie into a business portrait. Facial features are preserved. We don't support face swapping or creating images that mislead about real people.

How long does one edit take?

Usually from ten seconds to a minute — depending on the image size and the load on the service.

What language should the request be in?

Plain English, in ordinary words — the way you'd explain the edit to a person.

Can I use the result commercially?

Yes. The one thing to watch for is third-party logos and trademarks, whether they were in the source photo or appeared in the result.

What if the edit didn't come out right?

Run it again: every generation is independent. If it's consistently off, sharpen the wording — what to change, what to keep, and how the scene should be lit.