Generate images

Use @image to make an image from a description, or to change pictures you add. It’s good for concept art, reference sheets, texture ideas and the opening frame of a video.

Make an image

1. Click Image in the composer, or type @ and pick @image under Capabilities. An @image chip goes into your message, and the send button reads Generate image.
2. Describe the image: the subject, style, lighting and framing.
3. Check the price next to the send button, like Uses 1 prompt, then click Generate image.

Your words go to the image model as you wrote them. The agent doesn’t run, rewrite the prompt or touch your scene. A message makes one image, so send a new @image message for the next one.

Typing @image followed by a space also makes the chip. To turn image generation off, click Image again or delete the chip. A message uses one generator: turning on Image removes a @video chip.

Your prompt and any reference pictures are sent to a media generation provider to make the image. See the Privacy Policy .

Choose an image model

The Model menu in the image settings picks the model. Each option shows how many prompts it uses.

To put a person or a face in a video, make the picture with Seedream 5.0 from a description alone. Images of people made with GPT Image 2 don’t work in Seedance.

3D-Agent remembers your choices on your computer. When you switch to a model that can’t use a setting, the setting changes and the image settings say what changed, like Aspect ratio set to Auto.

Start from your own pictures

Add up to 3 pictures under References to edit them or to guide the new image.

1. Click Add reference.
2. Click Upload, drop a file in, or pick an image from this chat.
3. Describe what to change, like make this render look like a watercolor painting.

Reference pictures can be PNG, JPG or WebP, up to 12 MB each and 30 MB in all. The composer refuses a file it can’t take when you add it and says why. With two or more, each picture is numbered and used in that order. Drag a picture to move it, or select it with Tab and press ← or →.

To name a reference in your prompt, type @ and pick it, like @hero. Its name comes from the file name. To rename it, click its name under References. A name can have up to 32 lowercase letters, numbers, - or _.

Files attached to the chat aren’t sent to the image model. If your message has attachments, the composer shows Image and Video don't use chat attachments. Remove them, or add the pictures in the slots above.

How long can the prompt be?

An @image message can have up to 2,000 characters. The count shows at the top right of the message box, like 1,240 / 2,000 characters. Over the limit, you can’t send it, and the composer says how many characters to cut.

Use the image you made

The image appears in the chat with a progress bar and the time left while it’s made. Hover its caption to see the file name. Under the finished image, each chip readies your next message, and nothing is sent until you send it:

• Chat adds the image to the chat, for the agent to use in Blender.
• @image makes the image a reference for your next @image message, to edit it.
• @video uses the image in your next video, as a Reference, the First frame or the Last frame. See Generate videos.

Save as…, next to the caption, saves a copy wherever you choose.

Click the image to see it full size, with Edit in next message and Save as…. Press Esc or click Close to close it.

A Seedream 5.0 image made from words alone, without reference pictures, shows Works for people in Seedance. Only images like that work for a person or a face in a video. Seedance can’t use photos of real people, and images of people made with GPT Image 2 don’t work in Seedance either.

Where images are saved

Each image is saved as a PNG in the chat’s own folder. Library in the sidebar shows the images and video clips from every chat, with the settings they were made with. On a Library card:

• Use as reference adds the file to the message you’re writing.
• Open chat opens the chat it was made in.
• Save as… saves a copy.

What does an image cost?

Each image uses prompts from your monthly plan. The number depends on the model and the resolution, and the composer shows it next to the send button before you send. If no image is made, the prompts go back to your plan, and 3D-Agent says so, like 1 prompt is back in your plan.

If an image isn’t made

The image card says why. Do what its second line says:

Still stuck? Get help and include the card’s text.

Next steps