generate-video-by-seedance

Doubao Seedance Video Generation

Use the script bundled with the skill to generate/edit videos via theByteDance Seedance video model on Volcano Engine Ark:

• Doubao Seedance series video models (via the Ark contents/generations/tasks API)
• Supports text-to-video and image-to-video (single-image / multi-image reference)

Default settings:

• Default model:doubao-seedance-1-5-pro-251215(can be overridden with --model )
• Default duration:5 seconds
• Default aspect ratio:16:9

Text-to-video

uv run {baseDir}/scripts/generate_video.py \
  --prompt "一个在草地上奔跑的小狗视频，阳光明媚，电影感" \
  --filename "奔跑小狗.mp4"

Using a YAML configuration file

You can write all video generation parameters into a YAML configuration file, then use it via the --config parameter:

Text-to-video configuration example (text2video.yaml)：

prompt: 一个在草地上奔跑的小狗视频，阳光明媚，电影感
filename: 奔跑小狗.mp4
ratio: 16:9
duration: 5

Image-to-video configuration example (with a list of URL reference images,image2video.yaml）：

prompt: 将图1的表情换为图2的表情，做一个眨眼微笑的短视频
filename: 表情女孩.mp4
images:
  - https://ark-project.tos-cn-beijing.volces.com/doc_image/seedream4_imagesToimage_1.png
  - https://ark-project.tos-cn-beijing.volces.com/doc_image/seedream4_5_imagesToimage_2.png
ratio: 9:16
duration: 5

Local reference image configuration example (local_images.yaml）：

prompt: 多图融合成一个动画短片
filename: 多图融合动画.mp4
images:
  - /path/to/参考图1.jpg
  - /path/to/参考图2.png
ratio: 16:9
duration: 8

Usage:

uv run {baseDir}/scripts/generate_video.py --config image2video.yaml

Mix configuration files and command-line parameters (command-line parameters take higher priority):

uv run {baseDir}/scripts/generate_video.py --config prompt.yaml --filename "自定义文件名.mp4"

All parameters supported by the YAML configuration file:

• prompt: Video content description
• filename: Output file name
• images: List of reference images (can be URLs or local file paths)
• ratio: Aspect ratio (e.g., "16:9", "9:16")
• duration: Video duration (seconds)
• model: Full model name (advanced option)
• api_key: API key

Image-to-video / reference images (URLs or local files are both fine, multiple images, uses the lite i2v model)

uv run {baseDir}/scripts/generate_video.py \
  --prompt "参考图片中的女孩，做一个挥手微笑的短视频" \
  --filename "挥手女孩.mp4" \
  -i "https://example.com/ref_image_1.png" \
  -i "/path/to/local_ref_2.jpg" \
  --ratio "9:16" \
  --duration 8

Reference images only (no explicit text prompt)

uv run {baseDir}/scripts/generate_video.py \
  --filename "仅参考图生成视频.mp4" \
  -i "/path/to/ref1.jpg" \
  -i "/path/to/ref2.png"

Tip: At least one of "prompt" or "reference image" must be present, otherwise the script will report an error.

Notes

API key

• You need to set the ARK_API_KEY environment variable

Parameter descriptions

• prompt: Video content text (Chinese / English both fine)
• image / -i：
  • URL: Passed through directly to Ark (e.g., a publicly accessible image link)
  • Local file: Automatically read and converted to data:image/...;base64,... form before sending
  • Can be passed multiple times to form a multi-image reference (multi-image currently only takes effect under doubao-seedance-1-0-lite-i2v-250428 and other lite i2v models)
• ratio: Aspect ratio, commonly used:
  • "16:9"(default)
  • "9:16"
  • "1:1"
  • "21:9"
• duration: Video duration (seconds), determined by the range actually supported by the model / Endpoint
• model(optional, advanced): Directly specify the Ark Endpoint / model ID, overriding the default binding

Polling script get_video_task_status.py extra parameters:

• interval: Polling interval (seconds), default 10, can be adjusted as needed based on task duration and frequency requirements
• timeout: Maximum wait time (seconds), default 600, after which the script will report an error and exit

Results and file output

• Ark video generation is usually anasynchronous task：
  • The script will call POST /contents/generations/tasks
  • On success, it prioritizes parsing the returned id / task_id as the task ID
  • If the response body directly carries video_url, the script will try to fetch and save the video locally
• Save path rules:
  • If --filename If no directory is specified, it is saved by default to the outputs/ directory under this repository
  • If the filename has no extension, it is appended by default with .mp4

If the call succeeds but the backend only returns a task ID and has not yet generated video_url, the script will output the task ID, which can be used together with a polling script:

# 1) 先创建任务（可能只返回 TASK_ID）
uv run {baseDir}/scripts/generate_video.py \
  --prompt "一个示例视频" \
  --filename "示例视频.mp4"

# 2) 拿到上一步输出的 TASK_ID 后，轮询直到生成完成并下载视频
uv run {baseDir}/scripts/get_video_task_status.py \
  cgt-20260226184301-4h8v6 \
  --filename "示例视频.mp4" \
  --interval 5

The polling script will:

• Periodically call GET /contents/generations/tasks/{task_id} to check the task status
• When the status is succeeded / completed and video_url is obtained, automatically download the video locally

Filename recommendations (for caller reference)

• Do not include specific implementation details in the filename (such as "seedance", "ark", etc.)
• The filename language is recommended to match the Prompt language; for example, use a short Chinese filename for a Chinese Prompt (such as 挥手女孩.mp4）
• The filename should be meaningful but as short as possible, avoiding overly long sentences

Keyword hints (for discovery / search hits)

• "doubao", "豆包", "Seedance"
• "ByteDance video model", "字节跳动视频生成"
• "文生视频", "图生视频", "视频生成", "短视频生成"