青峰智影青峰智影
API ReferenceMirava
AI Model APIVideosSeedance Format

Request Parameters

metadata generation parameters, supported resolution and duration ranges, and reference asset count limits.

metadata Parameters

ParameterTypeDescription
resolutionstringOutput resolution, e.g. 480p / 720p / 1080p
durationnumberVideo duration (seconds)
ratiostringAspect ratio, e.g. 16:9 / 9:16 / 1:1 / 4:3 / 3:4 / adaptive
seednumberRandom seed; the same seed improves result consistency
fpsnumberFrame rate
generate_audiobooleanWhether to generate audio
enhancebooleanWhether to enable quality enhancement
return_last_framebooleanWhether to return the last frame image
watermarkbooleanWhether to add a watermark
reference_modestringReference mode, controlling how reference assets are used
audio_urlsarrayList of audio URLs

Supported Resolutions and Durations

ModelResolutionDuration (seconds)
doubao-seedance-2-0-260128480p / 720p / 1080p5 / 10
doubao-seedance-2-0-fast-260128480p / 720p / 1080p5 / 10
doubao-seedance-2-0-mini-260615480p / 720p5
doubao-seedance-2-5-260628720p / 1080p5 / 8 / 10

Combinations outside the table above are rejected. 2.5 is stronger in framing control and reference consistency, while the 2.0 series has advantages in longer durations and cost.

Key Differences Between 2.0 and 2.5

Capability2.0 Series2.5
Audio generationSupportedSupported (better quality)
Reference asset combinationsSupportedMore combinations supported
Last frame returnSupportedSupported
Framing controlBasic ratiosSupports adaptive
Duration tiers5 / 105 / 8 / 10

Reference Asset Count Limits

TypeLimit
Images30
Videos10
Audio10
Total assets50
content array items51 (including text)

Assets in the asset library must match the group and channel family of the current api_key. Cross-group asset references are rejected.

How is this guide?