Skip to Content
DocsDeliveryA finished film

A finished film

A production can end at the screenplay or at the hint, or go on to a finished film. The film is either rendered directly in 3D from the same code, or generated by a video model from the hint. Both start from a performance already accepted on the coarse forms.

Direct 3D film

A direct 3D film is the programmed film itself, rendered at its commissioned detail. Motion, staging and camera are accepted on the coarse forms first. Each object and set then takes its detail option on the same entity, which keeps its accepted placement and the performance around it. People stay the polygon man.

The look is the one PROJECT.md records: on a series from the settings’ cinematography record, on a short from you. The film sets one exposure and tone curve among those the renderer offers, places each light at its motivated source, and keeps the grade in the film module.

The agent checks the beauty render at delivery size for these:

  • each subject matches its record and canon references;
  • each detailed object and set matches its coarse form;
  • light direction, color and exposure hold across each scene’s cuts;
  • contact shadows ground every figure and object;
  • consecutive frames show no aliasing, shimmering, flicker or popping;
  • no surfaces pass through each other or flicker where they meet.

Camera moves fast enough to judder are listed for you to watch, because stills cannot show judder.

Generated film

A video model turns the motion hint into the finished picture and, when it can, sound. The hint supplies the performance, and image references and prompts supply the look.

The agent chooses the model with you and records it in PROJECT.md before preparing any input that depends on it. The hint must have passed segment review, and any required screenplay review must be closed. Generation then runs in this order:

  1. Write the shared description of the film’s look.
  2. Prepare the canon references the model takes.
  3. Plan each scene’s requests within the model’s limits.
  4. Prepare the shot frames and control channels the requests need.
  5. Decide each request’s sound, if the commission includes sound.
  6. Write each request’s prompt.
  7. Submit the requests, inspect the takes and adopt the ones that keep the beat.
  8. Repair, upscale, conform and match the accepted takes.

Each request is recorded with its exact input bytes and source revision. Before batching, the agent proves one representative contact, performance and identity case through inputs, generation, download and extracted stills. It generates the recorded number of takes and chooses among them before changing a prompt. Takes are placed by their decoded timestamps, and each take’s in and out frames are chosen on its own frames, because a model keeps the hint’s timing only approximately.

Acceptance

A finished film is judged by its own picture: preserved action, identity, continuity and audience effect, not its conditioning. The agent inspects stills from the source and from the delivered file at matching times, covering the entire film and all its joins, and measures the file against the delivery profile in PROJECT.md.

The result goes into docs/acceptance/<film>.md, with what you watched or heard and what nobody did. Independent review lists what that record holds.

The handoff for a generated film also carries rights and attribution, with each generation request traceable to the derivatives the film uses, the takes that carry content credentials, and the disclosure the release owes.

Last updated on