Text-to-video
Also known as: generative video, video generation
Models that create footage from a written description, which is a different business from editing footage somebody actually filmed.
Updated
What it means
Text-to-video generates frames that were never recorded. A description goes in and a short clip comes out, synthesised rather than selected, which is the opposite of what a highlight pipeline does.
The distinction matters because the two get conflated constantly. Editing automation chooses among real moments; generation invents them, and the honesty of the result is a completely different question.
For footage somebody shot themselves the generative route is mostly irrelevant. Nobody wants a synthesised version of their own run, which is why an AI highlight generator is a selection tool and not a generator.
How it works in practice
Where it does appear is in the gaps around real footage: a title background, an abstract texture behind a Lower third, a filler frame nobody is meant to study.
Those uses are low risk because nothing is being claimed about reality. A generated gradient behind a caption misleads no one and replaces a stock library rather than a camera.
Treating generated material as clearly separate from recorded material is the practical rule. Mixed into a sequence of real clips it reads as uncanny long before anybody can say why, and no amount of Color grading makes it sit comfortably alongside a real shot.
What to watch out for
Physics is where generated action footage falls apart. Boards, water and limbs behave almost correctly, and almost is immediately visible to anybody who does the sport.
Continuity is the second wall. Generated clips do not share a camera, a light or a location, so they cannot be cut together the way real coverage can, and Color matching cannot rescue them.
There is also a straightforward honesty problem. Inserting invented footage into a record of somebody day is a misrepresentation, however good it looks in isolation, and a viewer who spots it stops trusting the rest of the video as well.
How RawClip handles it
Nothing in a RawClip highlight is generated. Every frame comes from the uploaded files, because the point of the product is to find the moments that happened rather than to manufacture ones that did not. Titles and music are the only added elements.