Inpaint: redo one area of the image
Redo one area of an image, and only that one. You draw the area, you say what happens there, you infer. The rest does not move, down to the pixel.
Inpaint at Axiris redoes one area of an image, and only that one. You draw the area with the brush, the rectangle or the ellipse, you write what should happen there, you infer: outside the area, not a single pixel changes, and that is measured. Four models are offered according to the move: placing something new, transforming what is there, editing while keeping the person, strengthening the texture. The move is done in the axiris.ai interface, in the Box page.
Inpaint: redo one area of the image
An image is almost right. A sign is blank, an object is missing, a hand is wrong, a title needs to catch fire. Redoing everything means losing what was right. Inpaint redoes the area, and only the area.
The move
You draw an area on the image: with the brush, as a rectangle or as an ellipse. The area becomes a box in its own right, with its handles: you stretch it, you move it, you pick it up again. You write what should happen there, not a description of the whole image. You infer. The result is added to the displayed render; you move on to the next area.
Two areas can follow each other on the same image without loss: the second edit starts from the result of the first, and the person, the pose, the light stay what they were.
A progressive, adjustable edge blends the border of the area into the image: outwards, inwards or centred on the line.
Four models, by what they do
The choice is made in the mask bar, by the move you want. No engine name to know.
Model 1: place. Something that was not there arrives in the area: an object, a candle, flames that fill letters while keeping their exact shape. This is the model that holds best when you need to create within a precise frame.
Model 2: transform. What is there changes from inside its own material: an engraved lettering that catches fire, a surface that becomes something else, without the shape moving.
Model 3: edit. The area shows a person, an outfit, a setting; the prompt changes what it shows, and everything not asked for stays: the face, the pose, the light, the clothing. This model also reads the references you have set, up to nine, to bring a specific person into the area.
Model 4: strengthen the texture. Nothing is invented. Skin, stone, fabric, metal gain relief in the area, at equal definition.
For models 1 and 2, three intentions: inpaint (create), refine (refine, with an adjustable strength), remove (remove).
What does not move
Outside the drawn area, zero pixels changed: that is what we measure, not what we promise. A background can be frozen; it then becomes a layer that the area can only lighten, never soil. An exact text placed on the image keeps the shape of its letters while the fire passes through them.
What is validated no longer moves. That is the Axiris thesis applied to retouching: Building the image by areas.
Where it is done
In the axiris.ai interface, Box page: the Box Director holds the board, the boxes, the references and the layouts you save. The mask and its four models are there. Nothing to download: the models and the recipe stay with Axiris, you receive the image.
For the retouching method in general: Retouching by area. To put this move into a production: Axiris for business.