kie-seedance-video - SKILL.md Agent Skill

name: kie-seedance-video description: Generate AI video with ByteDance Seedance via Kie.ai — text-to-video and image-to-video, with optional audio. Supports Seedance 2, 2-Fast, 1.5 Pro, and 1.0 Pro/Lite. Triggers on: kie seedance, seedance video, seedance, generate a video, text to video, image to video, animate this image, make an ai video. argument-hint: [video prompt] allowed-tools: Bash(python3:*), Read, Write, Glob, Grep user-invocable: true

Generate video with ByteDance Seedance through the Kie.ai jobs API.

Setup: needs KIE_API_KEY in ~/.claude/.env. Get one at https://kie.ai.

Models

Key	Model id	Modes	Audio	Notes
`seedance-2` (default)	`bytedance/seedance-2`	t2v + i2v	yes (on by default)	Flagship multimodal; 480p/720p/1080p; 4–15s
`seedance-2-fast`	`bytedance/seedance-2-fast`	t2v + i2v	yes	Faster; 480p/720p only
`1.5-pro`	`bytedance/seedance-1.5-pro`	t2v + i2v	optional (off)	Durations 4/8/12s
`1-pro-t2v` / `1-pro-i2v`	`bytedance/v1-pro-*`	t2v / i2v	no	Seedance 1.0 Pro; 5 or 10s
`1-pro-fast-i2v`	`bytedance/v1-pro-fast-image-to-video`	i2v	no	720p/1080p only
`1-lite-t2v` / `1-lite-i2v`	`bytedance/v1-lite-*`	t2v / i2v	no	Cheapest tier

Flow

Step 1: Parse the request

$ARGUMENTS is the video prompt. Detect whether the user gave a start image (path/URL) → that means image-to-video.

Step 2: Pick the model & settings (ask, or use defaults on "just do it")

Model: default seedance-2. Use a 1-lite-* for cheap drafts; 1.5-pro for a good audio/cost balance.
Mode: image-to-video if a start image was given (--image), else text-to-video. seedance-2 / 1.5-pro do both; the *-t2v/*-i2v ids are mode-locked.
Duration: seconds. seedance-2 = 4–15; 1.5-pro = 4/8/12; v1 = 5/10. The script snaps your number to the model's allowed set.
Resolution: 480p / 720p (default) / 1080p.
Aspect ratio: 16:9 (default), 9:16, 1:1, 4:3, 3:4, 21:9.
Audio: add --audio for seedance-2 / 1.5-pro (native sound).
Last frame (optional): --last-frame <url/path> for first+last-frame interpolation (seedance-2, lite-i2v).

Step 3: Craft the prompt

Describe motion, subject, camera moves, lighting, pace. Show it to the user before generating. For image-to-video, the prompt describes how the still should move.

Step 4: Generate

python3 ~/.claude/skills/kie-seedance-video/generate.py "<prompt>" \
  --model seedance-2 \
  [--image <url-or-path>] [--last-frame <url-or-path>] \
  --duration 5 --resolution 720p --aspect-ratio 16:9 \
  [--audio] [--camera-fixed] [--count 1] \
  --slug <slug>

Videos take ~1–5 min. Mention that cost scales with model, resolution, duration, and audio (Kie.ai bills per generation — exact price is on the Kie Market page).

Step 5: Present results

Output → ~/videos/seedance/<date>-<slug>/. Show the path and offer: another duration/resolution, switch to a cheaper/pricier model, add audio, or animate a different still.

Rules

Image-to-video models require --image; the script auto-uploads local files.
Don't bake on-screen text into the prompt — burn captions later (e.g. /submagic).
Default to 16:9 for YouTube, 9:16 for Shorts/Reels/TikTok.