The previz desk captures more than stills: DOTV's previz has an animation timeline under the stage — walk your characters, orbit the camera around them, press Space to preview, then record frame by frame into an mp4. The recording can land on the canvas as a video node, ready to be used as a "generate from this" reference. This section covers how to lay down tracks, how one sentence can block the whole scene, segmented camera moves and safe follow shots, previewing and exporting.
The timeline is open by default (the Animation timeline button on the toolbar, or T, collapses it), and the seam on its top edge drags to resize. Each lane is one object's motion; every dot on a lane is a keyframe — where that object is, and in what pose, at that moment.
The group on the right is for watching: play / pause (Space), Loop, speed (0.25×–4×, preview only — the export always records frame by frame), and the total duration in the seconds field at the far right (0.5–120 s). Dragging the playhead only looks; editing the duration changes how long the piece is.
Select an object in the scene tree or the viewport first — that is what enables Add path at the left end of the timeline. Four shapes:
| Shape | What it does | When to use it |
|---|---|---|
| Line | Walks a straight line from where it stands (it turns to face the direction) | Blocking: from A to B |
| Circle | A loop around its current spot | Turning in place, a showcase orbit |
| Rectangle | A circuit along a rectangle | Patrolling, pacing back and forth |
| Draw | Draw the path yourself in the viewport, then press Done | Routes no regular shape can describe |
Once a track exists, refine it: drag keyframes along the lane to change when they happen; with the track selected, press Add a keyframe at the playhead (or I). Each track's row menu also has: Action (walk / run / the pose presets — none means it just slides), Face along the path, Hide, and Delete.
Cameras are a special kind of object: with a camera selected, an Add follow button appears next to "Add path" — pick a person or a prop and the camera keeps its eyes on it all scene long. The follow style lives in that track's menu: Orbit, Chase (from behind), Side (tracking alongside).
One motion for a whole track only ever says "he runs across". To see a door kicked in, a shot fired, someone pulled in behind you, you break the track into beats:
Every joint interpolates between two keys — so "start: aim, at 1.2 s: fire, land on: take cover" is three beats and an action you can actually read, instead of one pose sliding past. Follow the track clears the pose off this key and hands it back to the track's own motion.
For a character holding a firearm the same menu also offers Fire a shot here: at that instant the muzzle flashes, the surroundings light up and the weapon kicks. Mark a run of keys and it reads as automatic fire. On empty hands it does nothing — the flash lives on the weapon, so give them one on the right first (see 8.4).
One-sentence generation goes through the text model, the same pipeline and the same credit rules as one-sentence scene building (see 14.8).
Beyond hand-laying tracks, the left end of the timeline holds an input box: write "she walks to the window, camera pushes in with her" and the model stages the whole scene into a timeline — who goes where, doing what, how the camera follows — then fine-tune track by track.
Name the objects as the scene tree names them so the model can tell who is who. The result is a new timeline; when it can't stage your sentence, it says so and leaves your current timeline untouched — not one keyframe of what you built gets overwritten.
One shot does not need one camera move. Name the time periods and let the same shot push in, hold, then orbit or rise:
| Effect | Something you can say | What to verify |
|---|---|---|
| Segmented push | “From 0–2 seconds, push in slowly; hold at 2–3 seconds; from 3–6 seconds, orbit right and rise.” | Are each start and end exact, and does the frame genuinely hold still? |
| Follow without crowding | “Follow character A; tighten the distance through the doorway, then open it up outside.” | Is the distance smooth, and does the person stay in frame? |
| Separate head and body | “She looks toward the window first, then stands and walks to the door at 2 seconds.” | Are head, body and feet staged separately instead of sliding? |
| Seated dialogue | “This is a seated conversation; they may turn and gesture, but should not stand up midway.” | Does the seated pose hold, and do hands pass through the table or another person? |
Add keyframes to a follow track to change distance, angle or height at different times. Afterward, slow the preview down at transitions—especially doorways, turns and sitting down, where movement can suddenly speed up.
DOTV also flags common issues in the previz result: a camera too close to a wall, furniture or person; a character leaving frame; a prop passing through a wall; or a seated character changed to standing. Open the shot, look at the frame, then change the track, the placement or just that keyframe's action. The check is not replacing your taste; it surfaces problems that would be obvious in the final video.
Space plays and pauses; ← / → steps one frame (hold ⇧ for ten). What you see in the preview and what the export records frame by frame are necessarily the same picture; there is no “it looked fine but recorded wrong”.
Use Loop and speed together: loop a slow-motion pass to inspect how two moves join up, or run the whole scene at 2× to check the pacing.
Captures (8.7) and exported video are the chapter's two exits: stills feed image nodes as composition references, motion feeds video nodes as a previz clip — any shot you can't put into words, stage it here and let it move.
Happy with it? Press Export video at the top right. Two destinations:
The line pre-filled in that draft is the shooting brief for the final generation. The previz conveys camera movement, shot size, cut timing, positions, motion direction, spatial and scale relationships; you add limb detail, expression, wardrobe, physical detail and final lighting. The video model re-renders the same camera and blocking as a realistic picture.
Recording takes about as long as the clip itself; the progress bar sits in the top bar and keeps going if you wander off. If a recording fails (say the frame breaks midway), the reason stays on the top bar — not a toast that vanishes — so you can see what happened when you come back.