Text-to-video
Text-to-video describes producing a moving scene from a written description alone, with no reference image. It is the fastest route to a first impression and the least reliable route to a particular result.
What it suits
The start of a project, while the direction is still open. Five descriptions give five worlds, and watching them shows which one carries. That is cheaper and more honest than a long argument about words. It is also useful for movement studies, to see how a model handles a particular camera move.
Why we work differently for production
A prompt never describes a scene completely. What is not stated is decided by the model, and decided anew on every run. For a film whose shots have to match each other, that is the wrong starting point. So we first produce a still, get it approved, and derive the movement from it. That is image-to-video.
The advantage is obvious: you see and approve what will be on screen instead of trusting a description. See styleframe and approval point.
Frequently asked questions
What is text-to-video?
Text-to-video describes methods where a generative model produces a moving scene from a text description, without an image being supplied. The term contrasts with image-to-video, where a still forms the basis.
How long are the produced scenes?
Depending on the model, usually a few seconds. Longer scenes come from joining them, which in turn demands consistency across the joins. That is why a film consists of many short shots rather than one long one.
Can you make a whole spot with it?
Technically yes, reliably no. What is missing is control over recognisability and continuity. How tool and production differ is set out in AI video or traditional shoot.
Related terms
Read on: All terms in the glossary • AI video or traditional shoot
See it first, produce it second
We work from approved stills. What that means for your project takes 30 minutes to explain.
Book a first call