Seedance 2 Playbook

What This Skill Is

This is a directing playbook, not a prompt dump.

Use it when the user needs help with:

• turning a concept into a short Seedance-friendly video brief
• deciding whether an idea fits Seedance 2.0
• writing structured prompts instead of keyword soup
• shaping short videos such as:
  • music videos
  • cinematic commercials
  • healing lifestyle shorts
  • action or transformation sequences
  • POV or performance-led clips

The goal is to make Seedance behave like a short-form video director.

What This Skill Is Not

Do not use this skill as the only plan when the task mainly needs:

• strict talking-head delivery with exact spoken copy
• precise lip sync to a locked script
• high-volume localized avatar production
• platform-specific UGC ad strategy (high-volume variant testing, retention hooks, proof sequences)

Those cases usually need a different stack. ugc-ads is better for performance creative strategy, and kling-avatar is better for exact lip-synced presenter control.

Single-person product demo clips (UGC product demo format) are supported — see use-case-families. The exclusion above is for platform-native ad strategy and high-volume variant production, not for the visual format itself.

Core Rule

Write shots, not adjectives.

The community prompt library is valuable because the good examples do not say only "cinematic" or "cool". They specify:

• shot timing
• camera behavior
• subject behavior
• sound behavior
• transformation or transition logic
• what must stay consistent

If the prompt reads like a mood board instead of a shooting plan, it is probably weak.

Workflow

0. Prepare the starting image (I2V workflows)

If the concept uses Image-to-Video, the starting image is the single biggest quality lever. A weak starting frame produces a weak video regardless of prompt quality.

Find a suitable image model and generate a high-quality first frame before writing the video prompt:

# Search for high-quality image generation models on Pica
pica model search "nano banana"
pica model search "flux"
pica model info <image-model-id>

These search terms target popular photorealistic image models. Actual availability depends on what is currently live — use pica model info to confirm capabilities before generating.

Use a real reference photo (Pinterest, brand lookbook, film still) as a style target. Match the color grade, lighting, and composition you want in the final video.

Read input-modes § Starting Image Quality for details.

1. Decide whether Seedance is the right hammer

Seedance is strong when the task is:

• high-concept short video
• fast visual escalation
• multi-shot micro-narrative
• audio-aware or beat-aware motion
• product or fashion direction with strong camera language

Read use-case-families first.

2. Pick the structure before writing prose

Choose one structure:

• continuous tension shot
• 3-shot short
• timed multi-scene arc
• transformation sequence
• product montage

Read prompt-architecture.

3. Build the prompt in layers

A good Seedance prompt usually has these layers:

• style and world
• subject identity
• timing blocks
• camera instructions
• action choreography
• sound or rhythm cues
• constraints

Read prompt-blocks.

4. Keep each time slice responsible for one thing

For a timed prompt such as 0-3s, 3-7s, 7-11s, 11-15s:

• each slice should have one main visual goal
• do not introduce a new hero, camera gimmick, and environment change at the same time
• climax belongs near the end, not in every segment

Overstuffed prompts usually collapse into sludge.

5. Prepare voice assets (if the video needs speech)

If the concept includes dialogue or voiceover, generate the audio before the video — Seedance uses reference_audio to drive lip movement and pacing.

Recommended voice generation tools (ranked by ease of realistic output):

1. MiniMax — easiest path to natural-sounding speech
2. ElevenLabs V2 — highest quality with the right voice setup
3. Resemble AI — best for voice cloning and enhancement
4. Qwen2-TTS — free open-source option

pica model search "minimax speech"
pica model search "voice"

Alternative: find a real video clip with the vocal quality you want. Upload it as reference_audio and write your own dialogue in the prompt — the model adopts the vocal character while generating new speech.

Read input-modes § Voice And Audio Preparation for rules and examples.

6. Research the current model contract before generating

Never guess the live model ID.

Use:

pica model search "seedance"
pica model info <model-id>

If the task might fit a different route better, search that too:

pica model search "avatar"
pica model search "veo"
pica model search "commercial video"

7. Draft cheap, then escalate

First drafts are for validating:

• shot logic
• concept clarity
• pace
• whether the video has a real focal subject

Do not spend premium runs before the structure works.

Use quick-start-templates for ready-made prompt shapes and parameter values.

8. Review the result like an editor

Ask:

• Does the video have a clear hero subject?
• Is each segment legible on first watch?
• Did the camera language help, or just add chaos?
• Is the climax actually saved for the end?
• Would a simpler prompt be stronger?

If the result is noisy, read failure-modes.

9. Extend if the video needs to be longer than 15 seconds

Seedance generates up to 15 seconds per clip. For longer videos, chain clips:

1. Feed the output clip back as a reference video
2. Prompt the model to extend with one new element (new angle, new dialogue topic, new product)
3. The model maintains continuity automatically

3 extensions (4 clips, ~60s total) is the practical ceiling before identity drift accumulates. Review each extension before chaining the next.

Read input-modes § Video Extension for the full pattern.

Default Mental Model

Think in this order:

1. use case
2. structure
3. shot plan
4. sound/rhythm
5. polish language

Not the reverse.

References

• use-case-families: which Seedance jobs are worth doing
• prompt-architecture: reliable short-video structures
• prompt-blocks: how to write layers that Seedance understands (camera types, lighting, positive phrasing, intensity)
• input-modes: I2V (+ starting image quality), reference video, multimodal, video extension, prompt-as-image bypass, and Seedance talking head
• quick-start-templates: copy-paste prompt templates, parameters, and model selection
• failure-modes: common prompt mistakes and how to fix them
• archetypes/music-video: MV-shaped short videos
• archetypes/cinematic-commercial: product and fashion-forward commercial structure
• archetypes/healing-lifestyle: calming lifestyle or ASMR shorts
• archetypes/action-sequence: fast escalation, duel, transformation, clash

Minimal Workflow Example

# 1. Discover the live model
pica model search "seedance"
pica model info <model-id>

# 2. Write a structured prompt
pica generate \
  --model <model-id> \
  --kind video_generation \
  --input '{
    "prompt": "[Style] Futuristic performance MV, neon haze, photoreal. [Subject] One vocalist in a reflective silver coat, stable face and silhouette across all shots. [00:00-00:04] Medium push-in, singer stands still under a lone spotlight as bass rises. [00:04-00:09] Close-up and side-angle alternation, head movement and hand gestures sync with beat. [00:09-00:15] Wide reveal, floor lights ignite, camera circles as the performance hits climax. [Sound] dark synth pulse, beat-synced cuts, no subtitles, no extra performers."
  }'

First Follow-Up Skills

After this playbook proves useful, the most sensible dedicated follow-ups are:

• music-video
• commercial-video

Do not create those by copying case prompts verbatim. Extract the reusable directing patterns first.