All models

Seedance 2.5

bytedance/seedance-2.5
ByteDance Video GenerationImage-to-VideoReference ImagesReference VideosReference AudioAudio GenerationMultimodal

ByteDance's next-generation video model with a unified multimodal reference-to-video architecture. Generates video from text, up to 30 reference images, 10 reference videos and 10 reference audio clips - including audio-only input. Supports first/last-frame image-to-video, video editing, video extension and adaptive aspect ratio, up to 30 seconds at 720p.

Quick start

# Inspect the price — a plain request returns the 402 challenge:
curl -i https://api.glianalabs.com/v1/infer \
  -H "content-type: application/json" \
  -d '{
    "model": "bytedance/seedance-2.5",
    "prompt": "<your prompt>"
  }'

# Pay + run in one step with the mppx CLI (create a wallet: npx mppx account create):
npx mppx https://api.glianalabs.com/v1/infer \
  -J '{"model": "bytedance/seedance-2.5", "prompt": "<your prompt>"}'

Examples

Parameters

Input
aspect_ratio string optional default: "adaptive"

Video aspect ratio. "adaptive" automatically matches the aspect ratio of the provided reference image or video when applicable. First/last-frame generation always uses "adaptive"; any supplied value is overridden.

camera_fixed boolean optional default: false

Whether to fix camera position. Not currently supported by the provider; has no effect.

duration number optional default: 5

Generated video duration in seconds, 4-30 (default 5). Billed per second at the resolution tier; a request that supplies reference_videos or reference_audios is billed at the higher video-input rate.

fps number optional default: 24

Frame rate (frames per second)

generate_audio boolean optional

Whether to generate audio with the video

image string optional

First-frame reference image (HTTP(S) URL or base64 data URI) for image-to-video Pass a public URL; to use a local file, upload it (free) via POST /v1/media (≤40MB) and pass the returned url.

last_frame_image string optional

Last-frame reference image (HTTP(S) URL or base64 data URI). Requires a first-frame image to also be given. Pass a public URL; to use a local file, upload it (free) via POST /v1/media (≤40MB) and pass the returned url.

output_format string optional default: "mp4"

Output video container format. "mp4" offers the best compatibility and smaller file size; "mov" preserves higher color fidelity for professional post-production workflows at the cost of a larger file.

prompt string optional

Text prompt describing the video to generate. Optional when at least one reference image, video, or audio clip is provided (Seedance 2.5 supports audio-only input).

reference_audios array optional

Reference audio clips (0-10, HTTP(S) URLs or base64 data:audio/... URIs). Supports audio-only input (no image or video required). Total duration of all audio clips must not exceed 30 seconds. Pass a public URL; to use a local file, upload it (free) via POST /v1/media (≤40MB) and pass the returned url.

reference_images array optional

Reference images (0-30, HTTP(S) URLs or base64 data URIs) to guide multimodal video generation, editing, or extension. Pass a public URL; to use a local file, upload it (free) via POST /v1/media (≤40MB) and pass the returned url.

reference_videos array optional

Reference videos (0-10, HTTP(S) URLs or base64 data URIs) for style/motion guidance, video editing, or video extension. Total duration of all reference videos must not exceed 30 seconds. Pass a public URL; to use a local file, upload it (free) via POST /v1/media (≤40MB) and pass the returned url.

resolution string optional default: "720p"

Video resolution

seed integer optional

Random seed. Passing this does not error, but reproducibility is not guaranteed and is not documented by the provider.

use_virtual_avatar boolean optional default: false

Route image reference inputs (image, reference_images, last_frame_image) through ByteDance's trusted virtual avatar asset library before generation. Intended for AI-generated/virtual character avatars that would otherwise be blocked by face or deepfake detection

watermark boolean optional default: false

Whether to add a watermark to the output video

Output
video: URL to the generated video