Shapes are a decision, not an afterthought

A picture that is the wrong shape for its place gets cropped, and the crop is rarely the part you wanted. AI Pics asks for the shape before it makes anything.

The eight#

Square, 1:1, for posts and avatars. Classic, 4:3, for slides and photos. Portrait, 3:4, for prints and profiles. Photo, 3:2, for a camera frame. Poster, 2:3, for posters and covers. Wide, 16:9, for video and screens. Tall, 9:16, for stories and reels. Cinema, 21:9, for banners.

Choose by the slot#

Start from where the picture will sit. A hero banner is wide or cinema. A phone story is tall. A profile picture is square. When you know the slot, you know the shape, and you do not have to crop later.

A model may support fewer#

Each provider says which shapes and how many pictures per run it supports. The Preview supports all eight and up to four pictures. A real model may support a subset, and the composer shows the choices a model can make.

Size#

The longest side is limited by the model. The Preview paints at up to 1,024 pixels on the long side. That is fine for placeholders and previews on a screen, and not for print.

Variations keep the shape#

Variations take the settings of a run, so they keep its shape. If you want the same picture in another shape, make a new run and change the shape, keeping the seed.

For layouts#

If you are building a layout, make the stand-ins in the shapes of the slots, and keep them in a collection named for the project. When the real pictures come, you swap them in place and nothing reflows.

A small rule#

Decide the shape first, the words second. A good sentence in the wrong shape is still the wrong picture.

An example#

Suppose a page needs a wide banner, a square avatar and a tall story card. Make three runs with the same words and the same seed, one per shape, and keep them in one collection. You get three pictures that clearly belong together, in the right shapes, and the history shows how you made them.

See the composer.