seedance-cli

CLI for Volcengine Doubao Seedance video generation (Seedance 2.5 by default, 2.0/1.x selectable), with an accompanying SKILL for Claude Code / AI agents.

Install

# Recommended:
uv tool install zjandrew-seedance-cli

# Or with pipx:
pipx install zjandrew-seedance-cli

# Companion SKILL:
npx skills add zjandrew/seedance-cli -g -y

The PyPI distribution is named zjandrew-seedance-cli because the bare seedance-cli name was already taken. After install you still invoke it as seedance-cli (the binary name is unchanged).

Local development:

git clone https://github.com/zjandrew/seedance-cli.git
cd seedance-cli
uv sync --all-extras
uv run seedance-cli --version

Configure

# Interactive wizard (creates ~/.seedance-cli/config.json, chmod 600):
seedance-cli config init

# Or env vars:
export ARK_API_KEY=...
export SEEDANCE_ENDPOINT=https://ark.cn-beijing.volces.com/api/v3   # optional

# Or programmatic:
seedance-cli config set api_key ...
seedance-cli config set endpoint https://...

Priority: CLI flag > env var > config file > built-in default.

Multiple profiles

seedance-cli config list
seedance-cli config add prod
seedance-cli config use prod
seedance-cli --profile prod generate -p "..."

Models

Since v1.1.0 the default model is Seedance 2.5. Same commands now hit doubao-seedance-2-5-260628 (different capabilities and pricing than 2.0). Pin a profile back with seedance-cli config set default_model 2.0, or per-invocation with -m 2.0.

-m/--model accepts an alias below or any full doubao-seedance-* id (unknown ids pass through for forward compatibility).

--seed, --frames, --camera-fixed and --service-tier flex are 1.x-only; the CLI rejects them on 2.x up front. On 2.5, first-frame / first+last-frame tasks (and explicit --task-type edit/extend) force ratio=adaptive — drop --ratio and the model aligns it to your input.

Usage

# Text → video
seedance-cli generate -p "a tabby cat yawning at the camera" --ratio 16:9 --duration 5 --out cat.mp4

# Image → video (first frame)
seedance-cli generate -p "girl smiles" --image start.png --duration 5 --out smile.mp4

# First + last frame
seedance-cli generate -p "360-degree pan" \
  --image first.png:first_frame --image last.png:last_frame \
  --duration 5 --out pan.mp4

# Multimodal reference (seedance 2.x; up to 30 images on 2.5)
seedance-cli generate -p "..." --image a.png --image b.png --image c.png --out combo.mp4

# Video edit / extend (seedance 2.x). Video inputs must be web URLs —
# the API rejects base64 video; reuse a previous task's video_url or upload to TOS/OSS
seedance-cli generate -p "repaint walls blue" --video "https://example.com/orig.mp4" \
  --duration 5 --out edited.mp4

# Seedance 2.5: declare the task type to fail fast (sync 4xx instead of a
# deferred rejection), and pick the mov container for edit/extend chains
seedance-cli generate -p "repaint walls blue" -m 2.5 --video "https://example.com/orig.mp4" \
  --task-type edit --output-format mov --out edited.mov

# Async + polling
seedance-cli generate -p "..." --async
seedance-cli task list --status running --status queued
seedance-cli task get cgt-2026-... --wait --out result.mp4

# Dry run (prints the request body, no API call)
seedance-cli generate -p "..." --dry-run

SKILL

skills/seedance/SKILL.md ships in this repo. Install for Claude Code:

npx skills add zjandrew/seedance-cli -g -y

Exit codes

License

MIT