Editorial visualizationSeedance 2.0
Multimodal generation built around combined text, image, audio, and video references.
- Best for
- Multi-reference direction
- Inputs
- Text, Images, Audio, Video
A growing reference for understanding model families, input types, audio capabilities, output pipelines, and the versions currently available in the UnblurVideo workspace.
Each guide separates published model-family capabilities from provider-specific limits and current product availability.
2 guides · More in research
Editorial visualizationMultimodal generation built around combined text, image, audio, and video references.
Editorial visualizationA general-purpose multimodal system with stereo sound and a separate 2K regeneration workflow.
Model names change quickly. These capability groups provide a more durable way to understand how a system fits into a creative workflow.
How the model receives creative context before generation begins.
What the model produces directly and which steps belong to a separate pipeline.
The settings that shape framing, timing, repeatability, and delivery.
Guide coverage and generator availability are independent. Use this table to find the right detail page or move directly to a model that is already deployed.
Multi-reference direction
Text, Images, Audio, Video
Joint audio + video
GuideReference-heavy short scenes
Text, First / last frame, Reference media, Audio
32 kHz stereo
GuideByteDance's flagship Seedance model with text and image input, up to 1080p output, six aspect ratios, and optional generated audio.
Text, image
Optional
AvailableA faster, lighter Seedance 2.0 variant for quick iterations at 480p or 720p, with the same formats and optional audio.
Text, image
Optional
Available“Guide” means the model has an editorial reference page. “Available” means the listed production version can be submitted in the UnblurVideo generator today.
A practical framework for looking past showcase clips and choosing a model for a repeatable workflow.
A model with more input types is not automatically a better choice. Text is often enough for open-ended ideation; an image is more useful when appearance or composition must persist; first and last frames matter when the transition itself is the brief. Multimodal reference systems become valuable when motion, sound, and visual identity must be directed together.
Look beyond a strong first frame. Temporal consistency, subject identity, object permanence, camera movement, and prompt adherence determine whether a clip survives beyond a short demo. The right model depends on whether the scene is a simple motion study, a character performance, or a multi-shot sequence with stricter continuity needs.
Resolution, frame rate, audio, and duration are frequently described at the model-family level but exposed differently by providers. Native generation, regeneration, interpolation, and upscaling are distinct stages. A useful comparison states which stage produces each result instead of collapsing the whole pipeline into one headline number.
A model name can cover several versions, endpoints, and provider configurations. Before building a workflow around one, confirm the exact model ID, supported controls, duration, reference limits, pricing, and output handling available in the product you intend to use. This directory keeps published capabilities separate from current UnblurVideo availability for that reason.
The generator lists only deployed model IDs and their current controls, duration, and estimated credit cost.
A concise vocabulary for reading model announcements, provider documentation, and generation settings.
How editorial coverage, production availability, and model-family claims are handled across this index.