Everything FLUX 3 Image can do

FLUX 3 Image is Black Forest Labs' image model. It sets quoted words onto surfaces, edits a picture from a plain sentence, reads up to ten pictures at once, and makes images in fifteen shapes, from 21:9 to 9:21, up to 4K.

Every picture below was made by the step under it. Change a placeholder and press Gen to make your own.

Five parts, in order

Write the prompt in five parts, in this order: who, what they're doing, where, the camera, the style. The action is the part most often left out, and without it the pose changes every run. The model fills in whatever the prompt leaves open, so every detail you name is one it can't pick for you.

To make a set, keep the subject and change only the camera: the angle, the lens, the distance. One iced coffee, three shots.

Words on the surface

Put every word in quotes, then say how much room it takes ("two thirds of the can's width") and how big each line is next to the others. Write the lines in reading order, top to bottom.

One can becomes a product line: an edit changes the flavor and quotes the words that stay. The new words take the can's curve and its light.

The same works on glass, at night, in gold leaf.

Edit by naming it

An edit has no brush: the sentence does the pointing. Name the thing by what it looks like, say what it becomes, and list what stays. Three edits of one street: a swap, new light, and a removal. For a removal, say what was underneath.

Each edit's Resolution follows its input, so it keeps the street's shape.

Ten pictures, one frame

A step can read up to ten pictures. Give each one a job: here one is the person, one the jacket, one the place. Then say whose light wins, so she looks photographed there rather than pasted in.

A picture can lend its look and nothing else. This edit takes the pink evening light of the rooftop and puts it on the street, and keeps every object where it was. Drop in any picture as the look.

Any shape, up to 4K

Read a picture at a new shape and the model redraws the scene to fill it instead of cropping. The wide underwater still at the top becomes a tall phone story, with more water above the diver.

Size sets how much fine detail holds. 0.75K and 1K suit drafts and most pages; 2K, used here, holds skin, clay and fingerprints; 4K is for print and close crops. Each step up costs more: 4K costs about twelve times 1K.

To build on any step, press Mix to copy its recipe into your own page.