Core architecture
How does SyntarEngine make sure the final edit matches what I approved?
Last updated August 23, 2026
By breaking the production into setups the way a crew does, rendering each shot from a keyframe the director has already approved, and sequencing the results against the approved storyboard. Approval happens per shot, so a shot that misses is retaken on its own while everything approved around it stays exactly as it was.
Two ways to get to a finished video
Prompt, render, judge, repeat. The creator supplies references and a prompt, the model returns a take, and anything not specified is decided by the model. What arrives is a proposal. If it is wrong, the prompt changes and the take is generated again.
Specify, then execute. The creator defines the look shot by shot, approves a governing keyframe for each, and the system produces work that follows those directives. What arrives is the thing that was approved.
As models generate longer single takes, the first approach hands the model more of the composition. Each extension of the generation window enlarges the share of decisions the creator did not make. SyntarEngine runs the second approach, and that distance widens as models improve.
The work is grouped into setups
The way a crew breaks down a shooting day. One location, one lighting configuration, the actors and props that scene needs. Every shot sharing that setup is completed before the production moves on.
Continuity holds inside a setup because the shots share a defined look, and across setups because both DNA layers compose into every shot regardless of which group it belongs to.
Every shot starts from a keyframe you approved
A prompt describes an intention and leaves the model to interpret it. A keyframe is the frame: lighting key, blocking, lens character, palette, the position of a product in shot, all settled and visible before any motion is composed.
Cora composes the governing keyframe for each shot from the brief, the storyboard, and your DNA layers. You review and approve it. Generation begins from a state you have already accepted.
Approve per shot. Retake per shot.
The director approves shot by shot. Approved shots are retained. Only the shots that missed are retaken, and the approved work around them is untouched.
That is the part worth understanding, because it is unusual. In most tools the unit you regenerate is the unit you generated, so one bad beat costs the whole take. Here a bad beat costs one shot.
Two ways to retake
Keep the shot, change the action. The retake starts from the frame the shot already has, so framing, cast, wardrobe and lighting carry over exactly. What changes is the performance: a different line reading, an adjusted movement, a longer hold. This covers most retakes, because the shot was usually framed correctly and something inside it was wrong.
Change the shot. Assign or compose a new governing keyframe and the shot is re-conceived. Framing, blocking, and lighting key can all move.
Either way the retake cuts cleanly against its neighbours. It inherits the same approved keyframe, the same setup, and the same DNA layers the original had, and when you keep the shot it begins on the identical frame, so the cut into it is the cut you already approved.
The platform assembles the result
Retained shots and retaken shots are sequenced against the storyboard you approved. Shot order, timing, and continuity are held by the system rather than reconstructed by hand.
Shooting order optimises for setups. Assembly order follows the storyboard. Holding both is the platform's job, which is why grouping by setup does not push an editing burden back onto the director.
The finished edit is recognisably the thing approved at the Storyboard gate.
How long can a Production run?
Finished runtime is a function of the process rather than of a model's generation window. An individual shot is bounded by what a model produces in one pass, as it is everywhere. Total runtime is not, because the platform composes and sequences shots rather than asking one generation to carry the whole piece. Longer work means more setups.
Where this fits
SyntarEngine is built for work where the output has to match an approved intention: commercial and brand work that clears a client, a brand team, and in regulated categories a legal one.
Work where the goal is to explore quickly and accept a good result when it appears is well served by prompt-driven generation, and long single takes are genuinely faster for it. The pattern that approach produces is familiar though: generate a long take, dislike it, generate another, then another, and cherry-pick the usable pieces out of three of them by hand. The seconds that reach the edit are a fraction of the seconds paid for, and the assembly is manual.
Questions
What is a setup?
One location and lighting configuration, with the actors and props that scene requires. All shots sharing it are completed together before the production moves on, exactly as a crew would schedule a shooting day.
What happens when a shot comes back wrong?
It is retaken on its own. Everything approved around it is retained untouched.
Can I fix a performance without losing the framing?
Yes. Keep the shot and change the action, and framing, cast, wardrobe and lighting carry over exactly.
Do I have to reassemble the approved shots myself?
No. Edit, Finishing and Delivery are stages in the pipeline, so retained and retaken shots are sequenced against your approved storyboard without anything being exported for manual assembly.
Walk a Production through all six stages in the Open Set, then see the platform itself at syntarengine.com.