Seedance 2.0 JSON Schema for Structured Prompt Output: Complete Reference

Seedance 2.0 defines a strict JSON Schema in schemas/prompt-spec.schema.json that enforces 11 required fields—including sequence_relation, opening_state_source, and natural_language_prompt—to ensure every generation prompt contains complete sequencing metadata and role-based reference handling.

The Prompt Specification schema governs how all structured prompts must be formatted before submission to the Seedance generation engine. Located in the Emily2040/seedance-2.0 repository, this JSON Schema 2020-12 draft serves as the contract between client applications and the video generation pipeline, standardizing everything from clip sequencing logic to exclusion management.

Core Schema Requirements

Every valid prompt must satisfy the schema defined in [schemas/prompt-spec.schema.json](https://github.com/Emily2040/seedance-2.0/blob/main/schemas/prompt-spec.schema.json). The root type is object with no optional fields at the top level.

Required Fields

Field	Type	Description
project_id	string	Unique identifier for the parent project
clip_id	string	Unique identifier for this specific clip
prompt_version	string	Semantic version of the prompt schema (e.g., 2.0.0)
sequence_relation	enum	Relationship to preceding clips: standalone, sequence_first_clip, seamless_continuation, intentional_next_shot, bridge_between_known_states, repair_tail, reanchor_after_drift
generation_mode	string	Processing mode: typically standard or draft
reference_roles	array[string]	Roles providing reference material (e.g., director, cinematographer)
opening_state_source	enum	How the initial frame state is determined: planned_start_state, observed_end_state, user_supplied_final_frame, source_clip
current_clip_action	string	Description of intended action for this clip
endpoint	string	Target generation endpoint URL or identifier
completed_beat_exclusions	array[string]	Beat identifiers that must not be regenerated
reserved_future_exclusions	array[string]	Beat identifiers reserved for future generation
natural_language_prompt	string (minLength: 1)	Human-readable generation instructions

The schema enforces minLength: 1 on natural_language_prompt to prevent empty submissions. All arrays contain homogeneous string items with no additional constraints on the values themselves.

Sequence Relation Enum Values

The sequence_relation field controls how Seedance 2.0 handles temporal continuity between clips:

• standalone — Independent clip with no sequential dependencies
• sequence_first_clip — Opening clip of a multi-clip sequence
• seamless_continuation — Strict continuity from previous clip's end state
• intentional_next_shot — Deliberate cut or perspective change
• bridge_between_known_states — Interstitial clip connecting two established states
• repair_tail — Corrective generation replacing a failed previous ending
• reanchor_after_drift — Recovery generation when sequence coherence degrades

Opening State Source Options

The opening_state_source enum determines frame initialization:

Value	Use Case
planned_start_state	Pre-defined storyboard or shot list state
observed_end_state	Actual rendered output from previous clip
user_supplied_final_frame	Custom reference image provided by user
source_clip	Direct extraction from existing source material

JSON Schema Examples

Minimal Valid Prompt (Standalone Clip)

{
  "project_id": "proj-1234",
  "clip_id": "clip-01",
  "prompt_version": "2.0.0",
  "sequence_relation": "standalone",
  "generation_mode": "standard",
  "reference_roles": ["director", "cinematographer"],
  "opening_state_source": "planned_start_state",
  "current_clip_action": "establish_scene",
  "endpoint": "https://api.seedance.ai/v2/generate",
  "completed_beat_exclusions": [],
  "reserved_future_exclusions": [],
  "natural_language_prompt": "Create a wide-angle opening shot of a bustling city at dawn."
}

As implemented in Emily2040/seedance-2.0, this is the smallest payload that passes schema validation—all 11 fields present with valid types and non-empty natural language content.

Sequential Continuation (Multi-Clip Sequence)

{
  "project_id": "proj-1234",
  "clip_id": "clip-02",
  "prompt_version": "2.0.0",
  "sequence_relation": "seamless_continuation",
  "generation_mode": "standard",
  "reference_roles": ["director"],
  "opening_state_source": "observed_end_state",
  "current_clip_action": "follow_character",
  "endpoint": "https://api.seedance.ai/v2/generate",
  "completed_beat_exclusions": ["beat-01"],
  "reserved_future_exclusions": ["beat-04"],
  "natural_language_prompt": "Continue following the protagonist as they enter the subway, maintaining the established lighting."
}

This example demonstrates exclusion management: completed_beat_exclusions prevents regeneration of beat-01, while reserved_future_exclusions protects beat-04 for later manual creation.

Draft Mode with Custom State Source

{
  "project_id": "proj-5678",
  "clip_id": "clip-07",
  "prompt_version": "2.0.1",
  "sequence_relation": "intentional_next_shot",
  "generation_mode": "draft",
  "reference_roles": ["editor", "storyboard_artist"],
  "opening_state_source": "user_supplied_final_frame",
  "current_clip_action": "transition_to_climax",
  "endpoint": "https://api.seedance.ai/v2/generate",
  "completed_beat_exclusions": ["beat-03", "beat-05"],
  "reserved_future_exclusions": ["beat-09"],
  "natural_language_prompt": "Generate a high-energy chase sequence that picks up immediately after the previous beat, emphasizing motion blur."
}

The draft generation mode reduces processing cost for iteration, while intentional_next_shot signals a deliberate discontinuity from the previous clip's visual flow.

Schema Validation and Tooling

The Seedance 2.0 repository provides several files to enforce and test the prompt specification:

File	Purpose
schemas/prompt-spec.schema.json	Canonical schema definition
schemas/generation-run.schema.json	Parent schema wrapping one or more prompts
validation/fixtures/prompt-spec.valid.json	CI-tested valid example
scripts/prompt_lint.py	CLI validator for local development

Run validation locally against the JSON Schema:

python scripts/prompt_lint.py --schema schemas/prompt-spec.schema.json --prompt my-prompt.json

The lint script loads the schema and uses a JSON Schema validator to report any missing required fields or type violations.

Integration with Generation Runs

Individual prompts conforming to this schema are embedded within larger generation run payloads. The schemas/generation-run.schema.json file defines the container structure that batches multiple prompts for pipeline execution, ensuring each element passes the prompt specification before submission to the rendering engine.

Summary

• Seedance 2.0 structured prompt output requires exactly 11 fields defined in schemas/prompt-spec.schema.json
• The sequence_relation enum provides 7 values for continuity control, from standalone to reanchor_after_drift
• opening_state_source determines frame initialization: planned, observed, user-supplied, or source-derived
• natural_language_prompt enforces minLength: 1 to prevent empty submissions
• Exclusion arrays (completed_beat_exclusions, reserved_future_exclusions) manage beat-level generation scope
• Use scripts/prompt_lint.py for local validation against the canonical schema

Frequently Asked Questions

What version of JSON Schema does Seedance 2.0 use?

Seedance 2.0 uses JSON Schema 2020-12 draft as specified in the $schema declaration of schemas/prompt-spec.schema.json. This modern draft supports advanced validation features while maintaining broad tooling compatibility.

Can I submit a prompt with partial fields?

No. All 11 fields are required at the schema level—there are no optional top-level properties. Omitting any field causes validation to fail in scripts/prompt_lint.py and rejection by the generation pipeline.

How do I prevent specific beats from being regenerated?

Populate the completed_beat_exclusions array with string identifiers for beats that must be preserved. For beats you plan to create manually later, use reserved_future_exclusions. Both arrays accept any string format your project uses for beat identification.

Where can I find a working example of a valid prompt?

The file validation/fixtures/prompt-spec.valid.json contains a CI-tested valid prompt that passes schema validation. Use this as a starting template, or run python scripts/prompt_lint.py against your own payloads to verify compliance before API submission.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}