Everything Qwen-Image-2.1 can do

Qwen-Image-2.1 is the image model from Alibaba's Qwen team, here in its Pro version. It makes and edits pictures from a sentence, reads up to ten pictures at once, sets words in Chinese and English, and draws any shape from 1:8 to 8:1, up to about four megapixels.

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

Five pictures, one room

A step can read up to ten pictures, and none of them means anything until the prompt says so. Give each picture its job: here one is the empty room and four are catalog shots. Then place every piece against something already in the frame, "against the back wall", "at the right end of the sofa", and say what has to survive: each piece's color and shape, the room's window and light.

The four pieces are placeholders: drop in pictures of your own furniture and the same sentences lay them out.

Chinese and English, as written

Put every word to print in quotes, typed in the case you want it printed, and say where it sits and how it's set. A Chinese headline and its English line can share one layout.

Give each line of a poster its own place, size and weight. A quoted line with no place lands wherever there's room. Dashes and dots print as typed.

The shape the layout needs

Dimensions take any width and height from 1:8 to 8:1, up to about four megapixels, and every size costs the same. An 8:1 strip is a site header in one pass. Pick the shape first, then describe the frame from left to right: what sits at each end, what sits in the middle.

A tall frame reads the same way, top to bottom.

Empty space belongs in that order too. This 4:1 homepage hero gives its left third to bare wall, where the headline will go.

Say what stays

An edit has two halves: the change, and a roll call of what stays, written from what's in the frame: the sofa, the rug, the lamp, the print. "Keep everything else" names nothing. This one repaints the wall of the furnished room.

Each edit's Dimensions follow its input, so the room keeps its shape.

Words on a product change the same way. Quote the old words and the new ones, so the model knows which line to replace, and quote the lines that stay.

Every edit redraws the whole frame, and each pass leaves a little more grain. Make the changes you already know in one edit: this one swaps four pieces of clothing at once and keeps the face and the pose. Chain only the edits that need to see the last result first.

Rewriting, on or off

With Prompt rewriting on, a language model turns your prompt into a full scene before the picture is made: the props, the light, the hour, sometimes a sign or a name nobody asked for, so check a rewritten picture for words. It's a quick way to find a direction from a few words. Off, the model reads exactly what you wrote, and the rest of this page was made that way.

The rewrite is written fresh on every run, so the same prompt never brings the same picture back. Turn it off once you know what you want.

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