The form that works
The assistant looks for: subject, framing, light, materials, mood, style. Give it these six elements.Before / after
- Weak
- Improved
Mentioning a character
Type@Name to mention one of your characters (created in AI characters). Their reference images and their description go along with the request automatically. So write what they do, where and how, never their appearance.
- Weak
- Improved
Redoing or varying an image
Refer to the number: “redo number 3 as a tighter shot”, “same scene in daylight”. The assistant takes the image’s prompt, changes only what you ask, and attaches it as a reference if you want to keep the same subject. On an image, “Use as reference” attaches it to the next request, and “Animate” sends it to the Video clips tool.Common mistakes
- A request that is too short (“a kitchen”): add light, framing and style.
- Describing the appearance of a mentioned character: say only what they do.
- Putting readable text in the image (poster, sign): image models draw words badly. Add the text in the editor.
- Several requests at once without a number: “number 2”, “the last one” designate the session’s images, name them.
- An ambiguous request (which image, how many): the assistant asks you a question instead of launching.
Settings that go with it
Next to the “Request” field:- Engine: some engines accept more reference images or hold a face better. See Choosing an engine.
- Format, Resolution and Number of images.
- Attach: import an image (PNG, JPEG or WebP) or take an image from the session. If an engine doesn’t accept references, the app tells you before sending.
- The estimated cost is shown before sending. Above a certain threshold, the assistant asks for your confirmation with the amount: see Spending credits wisely.