01
Unified multimodal references
The official model supports text, image, audio, and video inputs in one generation architecture, allowing several types of context to shape a result.
AI video model guide
Seedance 2.0
A multimodal video model from ByteDance Seed built around text, image, audio, and video references with joint audio-video generation and detailed creative direction.
Published model guide
Model input mapByteDance Seed
Text
Images
Audio
Video
model contextgenerated result
Video output
This diagram explains supported context types. It is not presented as a generated sample from the model.
Capability overview
What defines Seedance 2.0
The points below summarize the model family as documented by its developer. Product-specific limits are kept separate.
02
Joint audio-video generation
Seedance 2.0 is designed to create motion and sound together instead of treating audio as a separate finishing step.
03
Director-level control
Reference materials and prompts can guide performance, lighting, shadow, and camera movement for more intentional scene direction.
Specifications
At a glance
Published model facts and current UnblurVideo status, last checked 2026-09-27.
- Developer
- ByteDance Seed
- Reference inputs
- Text, image, audio, and video
- Output
- Joint video and audio generation
- Creative controls
- Performance, lighting, shadow, and camera direction
- UnblurVideo availability
- PlannedNo fallback to another model is used.
- Facts last checked
- September 27, 2026
Best fit
Where Seedance 2.0 is most relevant
Its model design is most useful when the brief depends on several reference types or when picture and sound need to be planned together.
01
Multi-reference concepts
Combine visual, motion, and audio direction when a text prompt alone cannot communicate the full brief.
02
Performance-led scenes
Describe acting, staging, lighting, and camera movement as connected parts of the same shot.
03
Audio-aware storyboards
Explore scenes where dialogue, ambience, or sound cues are part of the concept from the beginning.
Workflow
A practical Seedance workflow
Plan the inputs first, then use the fewest references needed to communicate the shot clearly.
Define the shot
Write the subject, action, setting, camera movement, lighting, and intended sound as one concise direction.
Add purposeful references
Use images for appearance, video for movement, and audio only when each reference contributes information the prompt cannot express precisely.
Verify the deployed variant
Check the provider variant, accepted inputs, duration, resolution, and cost before generating; those product limits can differ from the model family overview.
Model availability and verified limits
This public guide is the canonical UnblurVideo page for Seedance 2.0 and clearly separates published model capabilities from the current product integration.
- This model guide is publicly indexable and ready to be expanded when the production integration is enabled.
- Seedance 2.0 is not in the production generation allowlist yet.
- The page does not silently substitute Seedance 1.5 Pro or another model.
- Exact duration, resolution, reference-count, and credit limits will be published from the deployed provider configuration when integration is complete.
FAQ
Questions about Seedance 2.0
Answers distinguish the published model family from the version and controls that may later be deployed here.
Related models
Continue comparing model families
Use the same availability and specification checks before choosing the next model for your workflow.
Production-ready options
Open generatorGenerate with a model available today
Open the shared workspace to compare the current production models, supported settings, duration, and credit cost before submitting a job.
Source for this guide: ByteDance Seed model page.