Pipeline steps

Script scan: your cast, world, and voice blocks

The script scan ("Analyze Script" on the Characters & Assets step) reads your script and produces two things: the project's assets — every recurring character, environment, and object, each with a written description — and the voice blocks, the narration and dialogue units the voiceover step renders one by one.

What the scan extracts

• ·Characters — every recurring figure in the script, with a description used to render its reference image. Speaking characters also get a voice binding later.
• ·Environments — the places scenes happen in.
• ·Objects — recurring items that need to look the same across shots.
• ·Voice blocks — the script split into narration and dialogue units, each attributed to its speaker.

Re-running the scan is destructive

Re-analyzing the script deletes and recreates all assets — there is no merging with the existing list. Reference images attached to the old assets go with them. Re-run the full scan when the cast itself has changed; otherwise use one of the non-destructive paths below.

After a script edit: re-extract voice blocks only

If you edited the script's wording but the cast and world are unchanged, use Re-extract from Script on the voiceover step. It re-runs only the voice blocks; assets and their reference images are left untouched.

Manual asset management

The scan coexists with hand editing. You can add an asset yourself (name and description, plus a voice for characters), edit any asset's description, or delete assets you don't want — without re-running anything.

Models

The scan model is user-pickable — current options include Gemini 2.5 Flash, GPT-5.4 Mini, GPT-5.4, and Claude Sonnet 4.6. Extraction runs at a deliberately low creative temperature internally, so results are consistent run to run.

Once the scan is done, the next task is giving every asset a reference image — that's what unlocks the voiceover step.