Seedance 2.0 Prompt Modes: Complete Guide to 7 Video Generation Modes and Their Use Cases

Seedance 2.0 provides seven distinct prompt modes—T2V, I2V, V2V, R2V, FLF2V, Edit, and Extend—that let creators generate, transform, or extend video content through a director-style intent interface.

Seedance 2.0 functions like a filmmaker's director: you describe what the scene should accomplish in natural language, and the skill compiles that intent into a production-ready prompt. The prompt modes define how Seedance interprets your input and what kind of video output to produce. This guide covers all seven modes, their specific capabilities, and when to apply each based on the Seedance 2.0 source code at Emily2040/seedance-2.0.

---

Seedance 2.0 Text-to-Video (T2V) Mode

T2V generates fresh video clips solely from text descriptions, with no source material required.

This is the foundational mode for concept-to-footage creation. You provide a natural language scene description, and Seedance renders it as a new video clip. According to docs/QUICKSTART.md, T2V serves as the default entry point for creators starting with only an idea.

from seedance import SeedanceClient

client = SeedanceClient(api_key="YOUR_KEY")

t2v = client.run_prompt(
    mode="t2v",
    prompt="A sunrise over a misty mountain range, cinematic, 4K"
)

Use T2V when you have no existing assets—just a vision for original content.

---

Seedance 2.0 Image-to-Video (I2V) Mode

I2V animates static images into motion clips while preserving the source visual style.

This mode excels at storyboard activation and concept art animation. Pass one or more images, and Seedance expands them into moving footage that maintains the original aesthetic. The schemas/prompt-spec.schema.json validates the image_path parameter as required for this mode.

i2v = client.run_prompt(
    mode="i2v",
    image_path="assets/concept_art.png",
    prompt="Animate the wind blowing through the trees in the picture"
)

Apply I2V to turn still photographs, illustrations, or storyboard panels into dynamic sequences.

---

Seedance 2.0 Video-to-Video (V2V) Mode

V2V re-renders existing video with modified parameters while conserving original motion.

This style transfer and quality refactoring mode takes your rough cut and produces a polished version with altered lighting, camera angles, or visual treatment. The original motion structure remains intact.

v2v = client.run_prompt(
    mode="v2v",
    video_path="assets/rough_cut.mp4",
    prompt="Re‑light the scene with golden hour lighting"
)

Use V2V for iterative refinement: applying new looks to locked edits without rebuilding from scratch.

---

Seedance 2.0 Reference-to-Video (R2V) Mode

R2V generates new footage that emulates the aesthetic of a provided reference clip.

This style emulation mode accepts a URL to a public video as a compositional and tonal guide. Seedance analyzes the reference's visual characteristics—color grading, framing, pacing—and applies them to your new content.

r2v = client.run_prompt(
    mode="r2v",
    reference_url="https://example.com/reference_clip.mp4",
    prompt="Create a new clip that matches the visual style of the reference"
)

Deploy R2V when you need to match a known creative reference or replicate specific shot composition.

---

Seedance 2.0 Full-Length-Film-to-Video (FLF2V) Mode

FLF2V produces extended cinematic sequences from high-level story outlines.

Unlike single-clip modes, FLF2V targets feature-scale generation. You provide an act-level or scene-level outline, and Seedance renders a cohesive, longer-form segment with cinematic quality.

flf2v = client.run_prompt(
    mode="flf2v",
    outline="Act 1: Hero wakes up in a neon‑lit city, discovers a hidden map.",
    prompt="Render the opening sequence as a cinematic short"
)

This mode suits pre-visualization, trailer assembly, and building extended narrative sequences from sparse story documents.

---

Seedance 2.0 Edit Mode

Edit performs targeted modifications to existing clips without full regeneration.

The in-place editing capability allows surgical changes: object replacement, lighting adjustments, mood shifts, or element removal/insertion. It preserves the base footage while altering specific attributes.

edit = client.run_prompt(
    mode="edit",
    video_path="assets/scene.mp4",
    prompt="Replace the red car with a blue one and brighten the sky"
)

Reach for Edit when you need rapid visual fixes or controlled alterations to approved footage.

---

Seedance 2.0 Extend Mode

Extend continues footage beyond its original endpoint with continuity constraints.

This temporal expansion mode analyzes the terminal frames of your clip and generates coherent follow-up action, maintaining camera movement, character positioning, and environmental consistency.

extend = client.run_prompt(
    mode="extend",
    video_path="assets/scene_end.mp4",
    prompt="Continue the action for another 5 seconds, keeping the camera movement"
)

Use Extend for pacing elongation, building longer continuous takes, or adding bridging shots between existing segments.

---

Seedance 2.0 Prompt Mode Architecture and Validation

The mode system is enforced through multiple layers in the codebase:

Component	Purpose	Location
Schema validation	Ensures mode field presence and valid values	schemas/prompt-spec.schema.json
Stress testing	Validates all modes across three prompt-writing styles	scripts/prompt_architecture_stress.py
Provider configuration	Selects model routing based on mode	agents/openai.yaml
Documentation	Lists modes with linked examples	docs/QUICKSTART.md

The stress test harness (scripts/prompt_architecture_stress.py) specifically evaluates each mode against:

• Director Formula — structured intent statements
• Listicle — enumerated description format
• Untrained — freeform natural language

This ensures robust handling regardless of how creators express their intent, as documented in RELEASE_v6.7.0.md.

---

Summary

• T2V — Create from scratch when no assets exist
• I2V — Animate static images preserving original style
• V2V — Refactor existing video with new visual parameters
• R2V — Generate footage matching an external reference aesthetic
• FLF2V — Build extended cinematic sequences from story outlines
• Edit — Modify specific elements within existing clips
• Extend — Lenghen footage while maintaining continuity

Each mode in Seedance 2.0 is accessible through the unified run_prompt() interface with mode-specific parameters validated against schemas/prompt-spec.schema.json.

---

Frequently Asked Questions

What is the default Seedance 2.0 prompt mode if none is specified?

The mode parameter is required per schemas/prompt-spec.schema.json. The API rejects requests without an explicit mode declaration. The Quick-Start guide at docs/QUICKSTART.md recommends beginning with T2V for first-time users, but no implicit fallback exists.

Can Seedance 2.0 prompt modes be combined in a single request?

No. Each run_prompt() call targets exactly one mode. For multi-mode workflows—such as generating a T2V clip then extending it—chain separate API calls. The examples/standalone-clip/prompt.md file demonstrates single-mode composition patterns.

How does Seedance 2.0 handle invalid or unsupported prompt modes?

Validation occurs at request time against schemas/prompt-spec.schema.json. Invalid mode values trigger a schema error before reaching the generation pipeline. The scripts/prompt_architecture_stress.py harness explicitly tests edge case mode inputs to ensure graceful rejection.

Which Seedance 2.0 prompt mode offers the most control over existing footage?

Edit provides the most granular control for modifications without regeneration. V2V offers broader stylistic overhauls while preserving motion. Choose Edit for targeted changes, V2V for comprehensive visual restyling.

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