Previs that directs the shot, not just describes it.
Trained to Imagine builds previs for film, television, and commercial productions: sequences blocked out in 3D so camera, lenses, framing, timing, and choreography are decided before the shoot. Unlike traditional previs, the blockout does not stop at planning — it can drive generative models directly, so the approved sequence becomes the structure for photoreal AI shots instead of a reference someone has to re-interpret. The team came out of feature VFX, nearly a decade at Marvel Studios, so the work is built to hand over cleanly to a shoot, an editor, or a VFX vendor.
A prompt can describe a car chase. A previs can direct one.
This sequence was designed as a Blender blockout first — camera, framing, timing, vehicle positions, and stunt choreography locked — and then generated photoreal, with separate references setting the vehicles, period, and look.
3D previs
Sequences blocked in 3D with real lens and camera data, so what is approved can actually be shot.
Previs to AI
The approved blockout drives generation, so the final shots follow the designed camera and timing instead of a model's guess.
Stunt and action design
Vehicle work, fights, and complex action choreographed and timed before anyone is on set.
Animatics and pitchvis
Cut sequences with temp sound for approvals, financing, and pitch — readable by people who do not read storyboards.
- 01
Script and boards
We start from the pages, boards, or a director's description of the sequence.
- 02
Blockout
Sets, characters, and vehicles roughed in 3D; camera, lenses, and timing worked out shot by shot.
- 03
Review and lock
Iterated with the director until the sequence is locked, cut as an animatic.
- 04
Hand over or generate
Delivered as camera data and reference for the shoot and VFX, or carried straight into AI-generated shots.
What is previs?
Previs, short for previsualization, is a rough 3D version of a sequence made before it is shot. It settles camera position, lens choice, framing, timing, and choreography in advance, so the shoot and the VFX work execute decisions that have already been made instead of discovering them on the day.
How is AI previs different from traditional previs?
Traditional previs ends as a reference: the shoot and VFX vendors rebuild everything from it. In a previs-to-AI workflow the same blockout conditions generative models, so camera movement, framing, and timing carry through into photoreal shots. That makes a sequence directed rather than described — the model executes a design instead of inventing one.
Can the previs be used for a live-action shoot?
Yes. Because it is built with real lens and camera data, the previs works as a shooting plan: the DP, stunt team, and VFX supervisor see the same sequence. AI generation is an option at the end, not a requirement.
What tools do you use for previs?
Blender for full sequence work, and Frameline 3D, the browser-based shot designer built by TTI Labs, for fast blocking, lens planning, and animatics. Output goes to whatever the next stage needs — editorial, camera data for VFX, or conditioning for generative models.
What are the limits of previs-to-AI today?
The hardest part is continuity: holding a vehicle's or character's identity and the spatial layout consistent across many shots. Single shots and short sequences work well now; long sequences need more passes and sometimes traditional VFX to finish. We say which applies at the start rather than after the work.