Operating sequence
Draft script-to-shot direction prompts from authorized beats, with narrative purpose, continuity, visual evidence, audio role, and rejection conditions.
- 01
Start every direction with a source pointer and narrative job
Use only an authorized script version. Begin the direction with the beat ID, the exact source span, what changes during the beat, and what the viewer must understand afterward. State which dialogue, action, factual claim, relationship, chronology, or emotional turn must survive. Mark interpretive choices and unresolved ambiguity instead of presenting them as source facts. Exclude confidential material that the next authorized step does not need. A good prompt is subordinate to the approved adaptation table: it cannot silently add motivation, evidence, products, events, or dialogue, and it does not establish that Seedance 2.5 can ingest the source or follow the wording.
- Beat and source pointer
- Viewer understanding
- Non-negotiable meaning
- Approved interpretation boundary
- 02
Specify one shot through observable visual and audio choices
Name the subject, action, setting, time, composition, camera height and movement, duration range, lighting purpose, and transition. Add the audio role: production dialogue, narration, ambience, effect, music landmark, or deliberate silence. Link authorized references by stable ID and describe what each reference governs. Carry character appearance, wardrobe, props, geography, screen direction, eyeline, and story state from the shot contract. Prefer concrete spatial and temporal language over mood adjectives. If a proposed visual is illustrative or composite, include the needed label. Do not invent controls, parameters, upload behavior, or accepted formats for an unverified product.
- 03
Create controlled variants and rejection questions
Draft variants by changing one editorial dimension—framing, pace, emphasis, or transition—while holding source meaning and continuity constant. For each, list disallowed outcomes such as omitted action, reversed causality, new factual implication, identity drift, prop mutation, geography conflict, unreadable evidence, dialogue mismatch, unsafe depiction, or missing disclosure. End with reviewer questions tied to the source and assembled cut: Did the beat land, did anything new become asserted, and does the next shot inherit the correct state? Log any later product test separately. These prompts support deliberate direction, but they do not demonstrate access, output quality, automation, repeatability, or successful adaptation.