Skill Details
Runtime
(claude-5-5-sonnet)
Example user message
[userMessage]: Seedance via Higgsfield, 9:16, one 10-second clip. Two characters, both have reference images attached. Maya (30s, long black braid, green rain jacket) and her brother Theo (20s, short curly hair, grey hoodie) sit in a parked car at night, rain on the windshield. English, neutral American accent. Script: Maya: 'You told me you'd quit.' Theo: 'I did. This is the last job.' Maya: 'It's never the last job.'
Example agent response
# Seedance Dialogue Prompt: "Last Job" (1 clip, 9:16, 10s)
## 1. Clip plan
| Clip | Duration | Speaker | Line | Words | Over limit? |
|---|---|---|---|---|---|
| 1 | 10s | Maya | "You told me you'd quit." | 5 | No |
| 1 | 10s | Theo | "I did. This is the last job." | 7 | No (two short sentences, under 10 words) |
| 1 | 10s | Maya | "It's never the last job." | 5 | No |
| **Clip 1 total** | | | | **17 / ~20** | Fits in one clip, no rewrite needed |
No lines were shortened or rewritten, so nothing is PENDING APPROVAL. The spoken words below are exactly as scripted.
## 2. Clip 1 prompt (ready to generate)
```
STYLE: Cinematic live-action drama, 9:16 vertical, shallow depth of field, cool night color palette, realistic and grounded. Duration 10 seconds.
SCENE: A sister confronts her brother in a parked car at night while rain falls on the windshield.
CHARACTERS (faces come from the attached reference images; do not alter or reinterpret them):
- Maya, 30s, long black braid, green rain jacket. Sits in the driver's seat. Reference image attached.
- Theo, 20s, short curly hair, grey hoodie. Sits in the passenger seat. Reference image attached.
LOCATION: Interior of a parked car at night, engine off, wipers off. Rain streaks and beads run down the windshield, with soft blurred streetlight glow outside. Dim, cool light inside the cabin. No other people.
CAMERA: Locked-off fixed medium two-shot from the dashboard side, both characters in frame, vertical 9:16. No cuts, no zooms, no camera movement at any point, especially during spoken lines.
DIALOGUE (time windows are labels; each line sits fully inside its window, with a short silence at the start and end of the clip):
- 0.0s to 0.8s: silence. Both characters still, mouths closed. Rain only.
- 0.8s to 2.8s: MAYA speaks, looking at Theo, quiet and controlled: "You told me you'd quit." Only Maya's mouth moves. Theo keeps his mouth closed and listens.
- 3.6s to 6.0s: THEO speaks, looking down then toward Maya, tired and guarded: "I did. This is the last job." Only Theo's mouth moves. Maya keeps her mouth closed and listens.
- 6.8s to 8.8s: MAYA speaks, steady and flat, holding his gaze: "It's never the last job." Only Maya's mouth moves. Theo keeps his mouth closed and listens.
- 8.8s to 10.0s: silence. Both characters still, mouths closed. Rain only.
LANGUAGE AND ACCENT: Maya speaks in English, neutral American accent. Theo speaks in English, neutral American accent.
DELIVERY: Calm, measured, deliberate enunciation. Low, quiet voices. No shouting, no crying.
AUDIO LOCK: These are the only words spoken in this clip. No ad-libbing, no extra speech, no narration, no voice-over, no subtitles, no captions, no on-screen text. No music. Quiet natural ambience only: soft rain on the glass and roof.
MOTION: Natural physics. Maya's braid hangs and shifts slightly with small head turns. Rain drips and streaks on the windshield. Jacket and hoodie fabric move only with small natural body movement.
LIP ARTICULATION: Natural, restrained mouth movement, clear closed-lip consonants, no exaggerated mouth shapes. A character who is not speaking keeps their mouth closed.
```
## 3. Test recommendation
Before running the full 10s clip, generate one isolated 5-second clip with a single line. Best choice: **Theo's "I did. This is the last job."** It is the longest line and the only one with two sentences, so it is the most likely to drift. Use the same two-shot, locks, and attribution, but with only Theo speaking and Maya silent. Check words, accent, and that only Theo's mouth moves.
If the single-line test shows the wrong person's mouth moving, use the cheaper fallback: split into separate clips that frame only the speaker, with the other person stated as off-screen and silent.
If you skip the test, a failed full-scene run costs several times as much.
## 4. Verification checklist (run after every generation)
- Transcribe or listen and compare word for word against the script.
- No added, dropped, or changed words. Log any mismatch.
- Order is correct: Maya, Theo, Maya.
- Right mouth on the right line. The listener's mouth stays closed.
- Accent is neutral American for both voices.
- No narration, music, subtitles, captions, or on-screen text.
- Camera stayed fixed, with no cut during a line.
- If anything fails, fix the cause (tighten a lock, shorten a line, or split the clip). Do not change the script to match what the model said unless the writer approves.
## 5. Assumptions
Tense, quiet tone (no shouting or sobbing) to keep mouth sync stable. Maya in the driver's seat, Theo in the passenger seat. Engine off, wipers off. Faces come from the reference images, so only hair and wardrobe are described. Results are not guaranteed: this makes exact words and the correct speaker far more likely and makes any miss diagnosable.
Get this skill + 9 more with SelectFrom $14/mo · Cancel anytimeFrom $14/mo · Cancel anytime
View SelectBy purchasing this skill, you agree to our terms of service
CLAUDE-5-5-SONNET
Turns scripted dialogue into Seedance-ready clip prompts that keep the spoken words exactly as written. Splits lines to the word limits, adds locks against ad-libbing, subtitles, wrong accents and lip-sync drift, flags over-long lines for approval, and gives a test-one-clip-first plan. Tested on Seedance 2.5 via Higgsfield.
...more
Added 1 week ago
