Turn images into a video

Pass one: what is this picture

Each upload is looked at on its own and comes back with a kind, a summary that names what it actually is, a transcription of every word visible in it, the concrete things on screen as short points, and any sequence evidence such as a step number or a breadcrumb. Its real pixel size is read from the file header rather than guessed, and turned into a shape of wide, square or tall.

Pass two: asset or instruction

This is the one people do not expect. Each image is also judged against your brief for whether it is a real asset to display or a reference you attached to guide the video. A mockup, a sketch, a screenshot of a written plan, a competitor's poster you want the style of, all come back as references. Those are pulled out of the picture pool, their transcribed content is folded into the brief as extra context, and they never appear on screen.

Pass three: what belongs together

A separate cheap pass looks across all the remaining images at once, because the per image analysis cannot see beyond its own picture. It returns groups with a short reason, such as steps of the checkout flow, and an index may only appear in one group. An image that stands alone is left out of every group. Grouped images are then shown in a single scene, in the order given.

One picture, one scene

After the script is written, a deduplication step runs over the scenes. If the director assigned the same image to more than one scene, it is kept in whichever scene uses the fewest images and removed from the others, along with that image's frame narration line and shape setting. A scene also uses either your uploads or generated illustrations, never both, so a picture is never sitting next to a fabricated one pretending to be from the same set.

How it works, in three steps

1. Step 1

   Say which pictures are references

   Write it in the brief. That is what the asset or instruction judgement reads, and it is the only way to keep a picture out of the video reliably.

2. Step 2

   Upload in the order they should play

   Display order is the index the director works with, and groups are played in the order they are listed.

3. Step 3

   Fix the shape per image if a crop looks wrong

   The inspector has a shape toggle per picture: the scene's own box, the biggest square inside it, or the biggest tall box.

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

Limits worth knowing

• Eight images per generation, up to 15 MB each, in PNG, JPG, WebP or GIF.
• Pictures are never upscaled, apart from a hero slot filling at least 60 percent of the canvas and even then by no more than 2.5 times.
• A scene either shows your uploads or generated images, never a mixture of both.
• Nothing is cropped to remove a background, a watermark or a face. The picture goes in as it arrives.

Questions people ask

Related pages