Use reference pictures and your own uploads

Some image models make a picture from your words alone. Others also work from pictures you give them: a person to keep, a product to show, a place or a look to follow. On Images, the Text to Image tab shows the pictures the model you've picked takes beside the words, and none for a model that reads words only, so pick the model first.

Add pictures

1. Pick a model that takes pictures. In the model menu, its row says how many, such as "10 image refs". A row that says "Words only" takes none.
2. Press Add, the dashed tile beside the words. A chooser opens with three tabs:
   • Uploads: every picture you've uploaded, in any tool, with your folders.
   • Made: the pictures you've made.
   • Saved: your Cast and your products.
3. Or bring in a new file. At the top of the chooser, drop pictures, paste one with ⌘V (Ctrl V on Windows), or choose files. New files are kept in Uploads, so they're there next time, in every tool.
4. Pick one or several and press Add.

You can also drop files straight onto a tile, or anywhere on the bar. Each picture shows as a spinning tile until it's up, and the line under the words says how many the model takes. The × on a picture takes it out.

Name them in the prompt

Each slot carries a tag in the order the pictures went in: @Image1, @Image2 and on. Type @ in the prompt to pick one from a list, and use the tags to say which picture is what, such as "@Image1 holds @Image2 up to the camera, warm window light". Naming them tells the model which picture is the person, which the product and which the place.

Change a picture you have

The Edit a picture tab changes a picture you give it rather than make a new one. Add the picture under Picture to edit, where it's @Image1, and say the change in the prompt; Generate stays off until it's in. Where the model takes more pictures, they go under More pictures as @Image2 and on: Nano Banana Edit takes up to 9 more, and Qwen Image Edit none. The tab's menu lists the models made to edit first, then the ones that also edit as they make.

To make a picture bigger, use the Upscale tab. To remove its background, relight it or erase something from it, use Edit & Enhance: see Upscale, cut out, relight and edit a picture.

Use a picture you made

Point at any picture in My images and press Use as reference. It goes straight in beside the words on Text to Image, and if the model you have picked reads words only, the bar moves to one that takes pictures. Edit this picture puts it on Edit a picture as the picture to change instead.

Once a picture is made, open it to see the pictures it was made from, under References beside its prompt, each by the name your prompt gave it. Press one to see it big.

Files

Pictures can be PNG, JPEG, WebP or GIF. A large photo is made smaller on your device before it uploads, so a big photo isn't turned away for its size. A GIF isn't resized, so it has to be under 8 MB.

When you change models

If you switch to a model that reads words only, the pictures you added stay in the bar, kept aside and unsent, until you pick a model that takes them. If you've added more pictures than the new model takes, Generate stays off and the bar says how many to take out.

Most models cost the same with pictures as without. Seedream 5.0 Pro adds a little for each picture past the first, and the price on Generate always includes it.

Elsewhere

• Video works from pictures too, as start and end frames or image refs: see Start and end frames, and reference clips.
• On a Creative Board, wire picture cards into an image card: each wire is a reference picture, up to what the card's model takes.

If a picture shows a real person, see Real people: faces, voices and consent.