AI images for video scenes

Only where the picture is generic

The rule the director works to is precise. Where a layout wants a photograph or an illustration and no uploaded image fits, it may write image prompts, one per image slot in slot order, describing a subject, a setting and a mood, with no text in the picture. It may only do that for generic visuals. It is told never to fake the user's own app, website, screenshots, real people or logos, and that for those it should prefer a layout that draws itself instead.

One or the other, never both

A scene uses either your uploaded images or generated ones, never a mixture. That prevents the specific ugliness of a real product photograph sitting next to a fabricated one at the same size, where a viewer reasonably assumes both are real. The same applies to the per frame narration, which is written the same way whichever kind of picture the scene is using.

The shape is chosen, not fixed

Every image slot publishes the box it makes in each canvas, and generation picks the aspect ratio nearest that box from eight options: 21:9, 16:9, 3:2, 4:3, 1:1, 3:4, 2:3 and 9:16. So a tall slot gets a tall picture rather than a square one cropped to fit. In the builder each media field also carries its own prompt box, up to 2,000 characters, so you can generate a replacement for one slot without touching the rest of the scene.

What generated images are actually good for

Backdrops behind a statement line, atmosphere in a hospitality or travel video, an abstract visual for a concept with no photograph, and a texture or setting shot to break up a run of typography. What they are bad for is anything a viewer will inspect: hands, faces, text on signage, a specific place they know. Video helps here, since a picture on screen for three seconds is examined far less than one on a web page.

How it works, in three steps

1. Step 1

   Upload what you have first

   Real pictures always win. Generation fills slots your own images do not, so eight good uploads leaves little for it to do.

2. Step 2

   Let the director write the prompts

   It writes one per slot in order, with subject, setting and mood, sized to the slot. That is usually better than a prompt written blind.

3. Step 3

   Regenerate one slot at a time

   Each media field in the inspector has its own describe and generate row, so a single weak picture can be replaced without redoing the scene.

The full walkthrough with screenshots is in the guide Make a promo video with AI voiceover.

Limits worth knowing

• Generated images are never used to depict your product, your interface, your logo or a real person.
• A scene uses uploaded images or generated ones, not both.
• The image prompt box takes 2,000 characters, and prompts are told to contain no text in the picture.
• Generated pictures land as ordinary attachments and count toward your account's storage like any other upload.

Questions people ask

Related pages