skl-seedance

name: skl-seedance description: Seedance 2.0 AI video generation — text-to-video and image-to-video via VolcEngine, fal.ai, or Replicate. Async submit → poll → download → vault note. No Docker, no MCP. Stores in research/videos/. skill_group: "ai-media" skill_category: "AI Image & Video Generation"

Seedance 2.0 AI Video Generation

Seedance 2.0 (by ByteDance) is a state-of-the-art video generation model available via multiple API providers. This skill covers the full generation workflow: submit request → poll for completion → download result → write vault note.

Prerequisites

Set one of the following environment variables based on your chosen provider:

Set provider: SEEDANCE_PROVIDER=fal (default) | volcengine | replicate

Operation: GENERATE

Submit a text-to-video or image-to-video request.

Required params:

• prompt ΓÇö text description of the video
• model ΓÇö model ID (default: fal-ai/seedance-1-lite for fal, adjust for other providers)
• duration ΓÇö video length in seconds (e.g., 5)
• aspect_ratio ΓÇö 16:9 | 9:16 | 1:1

Optional params:

• image_url ΓÇö seed image URL for image-to-video mode
• image_path ΓÇö local image path (fal provider uploads automatically)
• seed ΓÇö integer for reproducibility

Returns: task_id (string), estimated_cost_usd (float)

# httpx example (fal.ai provider)
import httpx, os

headers = {"Authorization": f"Key {os.environ['FAL_KEY']}", "Content-Type": "application/json"}
payload = {
    "prompt": "A serene mountain lake at sunrise, timelapse",
    "duration": 5,
    "aspect_ratio": "16:9"
}
r = httpx.post("https://queue.fal.run/fal-ai/seedance-1-lite", json=payload, headers=headers)
task_id = r.json()["request_id"]

Operation: STATUS

Poll generation status by task_id.

States: IN_QUEUE | IN_PROGRESS | COMPLETED | FAILED

r = httpx.get(f"https://queue.fal.run/fal-ai/seedance-1-lite/requests/{task_id}/status",
              headers=headers)
state = r.json()["status"]

Use exponential backoff: 2s, 4s, 8s, 16s, 32s, 32s… with a 300s wall-clock cap.

Operation: DOWNLOAD

Fetch the video from a COMPLETED job and save locally.

r = httpx.get(f"https://queue.fal.run/fal-ai/seedance-1-lite/requests/{task_id}",
              headers=headers)
video_url = r.json()["video"]["url"]

from datetime import datetime
from pathlib import Path
slug = prompt[:40].lower().replace(" ", "_").replace("/", "")
filename = f"seedance_{datetime.now().strftime('%Y%m%d')}_{slug}.mp4"
out_path = Path("research/videos") / filename
out_path.parent.mkdir(parents=True, exist_ok=True)
with httpx.stream("GET", video_url) as stream:
    with open(out_path, "wb") as f:
        for chunk in stream.iter_bytes(): f.write(chunk)

Operation: NOTE

Write an Obsidian-compatible vault note at research/videos/{slug}.md.

Required frontmatter:

---
date: YYYY-MM-DD
type: video
source: seedance
model: seedance-2.0
ingestion_type: ai-generated
title: "Prompt: {prompt[:60]}"
tags: [ai-video, seedance, generated]
tags: [video, seedance, generated, ai-video]
analysis_depth: generation_only
prompt: "{full prompt}"
duration: 5
aspect_ratio: "16:9"
output_file: research/videos/{filename}
provider: fal
created_date: YYYY-MM-DD
---

Operation: FULL

Convenience operation chaining GENERATE → STATUS (poll loop) → DOWNLOAD → NOTE.

agent: run FULL with prompt="..." [duration=5] [aspect_ratio=16:9] [image_url=...]

Prints progress: ⏳ Submitted task_id=xyz, 🔄 Polling… (12s), ✅ Downloaded: research/videos/...

Not In Scope

• Kling video models (use a separate skl-kling skill)
• Wan video, Mochi, CogVideoX (separate providers)
• Image generation (not video)

Error Handling