Everything Grok Imagine Image 2.0 can do

Grok Imagine Image 2.0 is the image model from SpaceXAI, the company that makes Grok. It sets words into a scene the way a designer would, keeps busy real-world scenes believable, and edits a picture from a sentence. It makes pictures at 1K or 2K, up to 3200 pixels on the long side, in thirteen shapes from 20:9 to 9:20. A new picture starts from words alone; an edit reads up to three pictures.

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

Thirteen shapes

Pick the shape the picture will be used at, from 20:9 to 9:20: the model composes for it instead of cropping a square. It also keeps space empty where you ask: the left third of the banner at the top for a headline, the top of this phone wallpaper for the clock. The two steps share one placeholder, so a new view fills both.

Five parts, finish last

Write the prompt in five parts: the subject, the setting, the composition, the light, and the finish, the line that says what kind of picture it is. The finish is the part most often left out, and without it the model picks the look. Here the first four parts stay the same and only the finish changes.

Words on every surface

Put each line of words in quotes and say what it sits on: the painted board, the enamel sign, the chalkboard, the life ring. This kiosk carries seven lines on five surfaces. End with "no other text", or the model fills empty space with words of its own.

Letters take on the material they're made of, so name the material and how it's made: stitched, bent in glass, pressed into paper. The three lines are the same words in each picture; only the surface changes.

On a layout, write the lines from largest to smallest and give each one a size against the others. The poster comes back in the order you wrote it: the band, the support act, the date, the venue, the small print.

For the next stop on the tour, an edit changes one line and keeps the rest: quote the new words and name what stays. Its Resolution follows the poster, so it keeps the poster's shape.

Name what has to be right

In a busy scene, list what a careful viewer would check: how many people and what they're doing, where the light falls, what the wet floor reflects. The model holds those steady instead of drifting to a vaguer scene. Keep the list to what matters most.

One change, the rest stays

An edit prompt is an instruction about the change, not a new description of the picture. Say what changes and name what stays. Three edits of one van: a new paint job, something taken away, a new place. For a removal, say what fills the gap.

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

Three pictures in one edit

An edit reads up to three pictures: the one it changes and two more. Say which picture is fixed and what comes from the other. Here the room stays exactly as it was, and the teapot moves in, lit by the room's own light. Make a different product above and press Gen here, or use the pencil to bring in a photo of your own.

Draft at 1K, finish at 2K

Quality is low or medium, and size is 1K or 2K. Settle the prompt at 1K and low, then make the keeper at 2K and medium, where fine texture and small letters hold up. A 2K medium picture costs twice a 1K low one, and an edit adds a little for each picture it reads. The two steps share one placeholder.

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