Want to make one of your own following someone else's film, or to lay out the structure of a final cut you already have? The quickest road is to have a model watch it once: DOTV's frame study breaks a video into four sets of references you can carry straight on with — shots, motion, music and script. This section covers what is on this card, what each of the four dimensions produces, where the output lands once it runs, and how the script is cut out one line per cut.
First wire in the clip it is to break down: connect a video node that has already generated (the one on the left in the picture). Once it is connected the card holds, in order:
| Dimension | What it produces |
|---|---|
| Shots | A set of storyboard images, one per shot, landing in a shots group |
| Motion | Camera-move clips, cut out shot by shot |
| Music | The score from this clip |
| Script | A script node holding this clip's shot list (see "Where the script comes from" below) |
All four are ticked by default — unticking one saves very few credits, because the cost is almost all in sampling frames (the model watches the whole clip either way, and it is the same pass whatever you ask it to break out).
The study runs in this page: lose the tab, or refresh, and the run is cut off. For a long clip, set the time aside and do not close the page halfway. On failure the card says the reason outright (the clip could not be read, the model did not answer, the result came back malformed), and the main button beside it becomes Study again: click it to run the whole thing over.
Nothing runs until you press Start study (this step is billed by sampled frames, so it is not pressed for you). While it runs the card is replaced by a placeholder panel with a pill saying where it has got to: Grabbing frames, Filling in the shot list 3/8, cutting motion clips, pulling the audio — those words follow the real progress. Meanwhile the dimension buttons are disabled and the main button spins; when it finishes, the four sets of output land to the right of this clip, numbered by shot.
The script is not the model grouping things by impression: the browser goes through this clip frame by frame first and finds the cut points from how the picture changes, one line per cut — so a fast cut every two or three seconds comes out too, instead of being merged into twenty-odd coarse shots the way the watch-through pass does. Each shot then has two frames taken, one earlier and one later, and handed to a vision model to fill in the picture, the shot size, the light and the camera move; the dialogue carries over from what the watch-through pass wrote down, split into the shots by time.
This pass happens in the browser and asks nothing of you in the way of installs. When no cut points can be found, or the vision model never answers, the script falls back to the rough draft from the watch-through pass — one fill-in failing does not fail the whole study.
If you have already run a study and simply find the script too coarse, there is no need to run it again (that would mean watching the clip again and adding a whole set of groups to the canvas): tell the Agent to re-split this script by cut points, and it replaces only the shot list inside the script node — it does not watch the clip, does not build groups, and keeps the assets, the look and anything you edited by hand.
The four sets of references are ordinary nodes: storyboard images can be edited and regenerated, motion clips can go on to be reshot, and the script node lays out a whole set of storyboards, a whole set of videos and a whole set of voiceovers in one click (see 6.1). What comes out of a study costs no credits — what you paid went on the run that produced it.