Mai Image 2.5 Edit: change one locked still (5cr)

Mai Image 2.5 Edit is an edit. Point at the object. A new text-to-image throws away the still you approved.

Mai Image 2.5 Edit is Microsoft's image-editing row: pixel-level cleanup, backgrounds, and text. Categories are edit-image and image-to-image. An image is required. The catalog lists 5 credits. Audio is off — this is a still. Aspect ratios include auto, 1:1, 4:3, 3:4, 16:9, 9:16, 3:2, and 2:3. There is no duration ladder because nothing here moves.

The mistake is regenerating the hero because a label is wrong. That is how you lose the lighting you already signed.

Point at the object

An edit names a region and a change. "Remove the cardboard behind the bottle." "Replace the shelf text with the SKU we actually ship." "Clean the sensor dust in the upper left." The rest of the frame is the reason you are here. If the instruction is "a premium product shot of our bottle in a kitchen," you have left the edit row and opened a lottery.

Mai Image 2.5 Edit is the Microsoft pass for that named change. It is not a second art director. It will not guess which bottle is the hero if you feed it a messy table and a poetic prompt. Crop toward the object, then point.

The AI photo editor is the Versely surface for this job. The model is the 5-credit row. Do not confuse the surface with a generate button that happens to accept an upload.

Cleanup is not a new render

Pixel-level cleanup, background, and text are three different notes. Stack them only if each note is small. A background swap plus a label rewrite plus a hand fix is three chances to drift. Do the background. Approve. Then the label. Approve. The still you already liked is the contract; each pass should leave it mostly alone.

Text-in-image is the note people try to "just regenerate." That is how the bottle becomes a different bottle that happens to say the right words. If the pack is approved, you are editing letters, not inventing a pack. Say so in the instruction. If the pack is not approved, you are on a text-to-image row, and this page is the wrong door.

Five credits is the listed cost of the edit. It is not cheaper than a generate in every catalog — it is cheaper than throwing away a still you already paid to lock. The expensive version of a label fix is the afternoon of identity retries after you treated edit as generate.

What you do not ask this row

Do not ask it to invent a scene. There is no video out. Audio is false. Durations do not exist. If the next step is motion, lock this still, then take it to an image-to-video model. Editing the still after you have already animated it is backwards: you will animate the defect, then pay to edit the defect in every frame.

Do not skip the upload. requires_image is true. A prompt without a plate is not this model. If you do not have a plate, you wanted text-to-image, not Mai.

Do not treat aspect ratio as a crop tool for a problem the pixels do not contain. 16:9 will not invent a wider room. If the room is missing, generate. If the bottle is wrong, stay.

The ranked edit map is best image editing model. This article is only the routing rule: change one locked still.

Circle one object

Look at the approved frame. Circle one object. Write one verb: remove, replace, clean, reletter. If you cannot circle it, generate. If you can, and you still open text-to-image "because it is faster," you are paying to lose the still. Use Mai Image 2.5 Edit. Then stop.

FAQ

Why not generate a new still if both are images?

Because a generate re-samples identity. The approved bottle, the approved light, the approved hand — all of that goes back into the draw. Mai Image 2.5 Edit is scoped to cleanup, background, and text on a supplied image. You keep the plate. That is the job.

Does this model require an upload?

Yes. The catalog marks requires_image true. There is no text-only path on this row. If you do not have a still, pick a text-to-image model. If you have a still and a named change, stay here.

Can I fix a talking clip with Mai Image 2.5 Edit?

No. This is an image row. Audio is off. There are no durations. Pull a frame if you must, edit the frame, and understand you are not editing the motion. Mouth work belongs on a lipsync row after the picture exists.

When do I stop editing and start over?

When the blocking is wrong — wrong angle, wrong room, wrong product. Pixel-level cleanup will not recompose a shot. Start over on a stills model, lock the new plate, and only then come back to a 5-credit edit for the last millimetre.