Text to video and image to video, side by side
A generation request usually takes one of two paths: a prompt alone, or a prompt with a supplied image to start from. The same page often publishes different shape rules, different conditions and sometimes different lengths for each. As of 2026-09-12.
| Path | What it starts from | What the page often says differently |
|---|---|---|
| Text to video | A prompt | A shape has to be chosen explicitly |
| Image to video | A supplied frame | The shape follows the input image |
| Either | Both are documented on one page | Length is often shared between them |
Inclusion rule. Input paths a generation page documents, with the kinds of difference those pages commonly state between them. Order. The simpler path first.
1The image path is how a cast stays recognisable
Serial and multi-shot advertising needs the same face across pieces that were generated separately. Supplying a frame is the most direct mechanism for that, which is why the image path is the normal one for anything with people in it.
It also decides the shape, because the input already has one. So on that path the shape question moves upstream to whatever produced the frame, and a widescreen reference cannot produce a vertical piece.
2Conditions differ more often than values
Where a page states a length once for both paths, a plan can mix them freely. Where it states a condition on one path only, mixing them means re-checking the arithmetic per shot.
The most common such condition is that frame conditioning and reference images cannot be combined in one call. That forces a choice per shot between pinning a composition and keeping a character reference.
3Which path a brief implies is worth settling early
An establishing beat with no cast in it comes from a prompt. A shot with a recurring character comes from a reference. A sequence usually needs both, and the shot list should say which is which.
That is not a technical detail. It decides where the shape is chosen, whether a condition applies, and which of the published limits on the page the request is actually held to.
A vocabulary entry: the definition, plus how the specification pages on these sheets usually publish it. Table checked 2026-09-12. Nearby entries: Reference image, Clip ceiling.
4Making the ad itself
For ad work around “Text and image paths”, SceneMixer's pages explain how its ad projects are produced:
Character sheets are always portrait and scene images always widescreen, whatever ratio the videos use.
SceneMixer, Vertical 9:16 or Landscape 16:9 — “Will choosing 9:16 make my character images vertical?” (read 2026-10-13)
The product image is always redrawn from your original photo so it stays your product.
SceneMixer, Ad Projects and Remake Projects: What Looks Different in the Wizard and the Editor, and Why — “Can I redraw the product from text?” (read 2026-10-13)