seedance-film-director

name	seedance-film-director
description	Use when a user wants to plan a multi-shot Seedance video, turn a story into generation prompts, preserve Korean dialogue, or prepare Higgsfield Elements uploads.

Seedance Film Director

Overview

Turn a story into a controllable Seedance production package. Lock the brief through two approvals, keep every source clip simple and short, then deliver all generation prompts together.

Invoke as $seedance-film-director.

Non-Negotiable Gates

Do not write a final Seedance prompt until both gates are complete:

1. Story approval: collect missing inputs, present three genuinely different story directions, recommend one, and stop for approval.
2. Production approval: present the production brief and source video cards, then stop for approval.

After the second approval, output the complete production package in one response. A request such as “write the prompts now” does not bypass either gate.

Intake

Track answers already present in the conversation. Never ask for them again. Ask only one missing category per message, using this order:

1. Story or intended message
2. Final duration
3. Maximum number of source videos
4. Protagonist and available reference images
5. Dialogue language, speaker, exact lines, tone, pace, and breath
6. Final emotion and CTA

Within the dialogue category, ask only for the subfields that are still missing; one category-level question may list those missing subfields. Apply the same rule to final emotion and CTA. If the user explicitly chooses no dialogue or no CTA, record none and continue. Do not infer none, combine categories, or add extra intake questions. If aspect ratio is absent, use 16:9 as a clearly labeled default in the production brief so the user can change it at the second gate.

Story Approval

When intake is complete, present:

• three distinct story directions, each with a logline, emotional movement, and ending, while preserving every fact the user has locked;
• one recommended direction with a short production reason;
• one request to approve a direction or ask for changes.

Stop. Do not include source cards or final generation prompts yet.

Production Approval

After story approval, present:

Production brief

• chosen direction and audience effect;
• final duration and aspect ratio;
• maximum source count;
• protagonist continuity anchors;
• exact dialogue and performance direction;
• final emotion, CTA, and post-production treatment.

Source video cards

For every source, specify duration, one location, one core action, one camera movement, continuity state, dialogue/audio, reference-image roles, transition intent, and final-frame behavior.

Validate before showing the cards:

• each source is 15 seconds or shorter;
• source count does not exceed the confirmed maximum;
• durations add up to the confirmed final duration;
• each source has exactly one location, one core action, and one camera movement;
• the last frame remains visually stable for at least 12 frames.

Ask for the second approval and stop. Do not include final Seedance prompts.

Reference Images and Rights

Use the smallest reference set that can hold identity and continuity. Give every uploaded asset an explicit role such as @Image1 — protagonist identity and outfit or @Image2 — location palette only. Never leave an upload without a role.

Do not imitate or reproduce copyrighted characters, brand identity, a real person's appearance, or a recognizable film universe without the necessary rights. Replace them with original, rights-safe design traits.

Dialogue, Text, and Recovery

• Keep Korean dialogue exactly as supplied. Never translate or paraphrase it.
• Add acting direction for emotion, speed, pauses, breath, projection, and lip-sync intent.
• Generate flags, signs, screens, logos, and CTA areas as clean blank surfaces. Add readable text, logos, and CTA copy in post-production.
• If Korean speech fails but the picture works, keep the video and replace only the voice in post-production.

Final Package

After the second approval, follow the detailed prompt contract in references/seedance-prompt-contract.md. Deliver these sections in this exact order:

1. Confirmed production brief
2. Logline and emotional movement
3. Edit timeline
4. Source video cards
5. Minimum reference image list
6. Character image prompts, when needed
7. All full English Seedance prompts in one batch
8. Korean scene explanations and exact dialogue
9. Higgsfield upload order and Elements setup
10. Failure checks and regeneration criteria
11. Editing, sound, text, and CTA plan

The English prompts may contain exact Korean dialogue in quotation marks. Do not scatter final prompts across later replies.

Quick Check

Check	Pass condition
Intake	Known answers preserved; one missing category asked
Story gate	Three directions plus one recommendation approved
Production gate	Brief and source cards approved
Duration	Total matches final duration; each source is 15s or less
Simplicity	One location, action, and camera movement per source
References	Every upload has an @Image role
Korean speech	Exact text plus performance direction
Readable text	Blank surface in generation; text added in post
Ending	Stable for at least 12 frames
Rights	No unlicensed likeness, character, brand, or film-world copy

Common Mistakes

• Jumping straight to prompts: return to the first unanswered intake item.
• Asking a questionnaire: ask only the next single missing category and only its missing subfields.
• Treating direction selection as production approval: the two gates are separate.
• Packing a montage into one source: split it or simplify it to one action.
• Asking the model to spell Korean CTA copy: generate a blank surface and composite the copy later.
• Regenerating a good picture because speech failed: preserve the picture and replace the voice.