Seedance

Seedance

Generate short videos with ByteDance Seedance via FAL — text-to-video or image-to-video.

This skill covers:

• Text-to-video (T2V) — generate a clip from prompt alone
• Image-to-video (I2V) — animate an existing still
• Pro tier (highest quality) and Lite tier (faster / cheaper)
• Optional ambient audio (Seedance audio-enabled endpoint)

This skill does not cover:

• Kling-driven motion with product-element conditioning — use /add-motion
• Talking-head / voiceover content — use /ugc or /product-ugc
• Long-form video (clips capped at 12s)

Output saved to ./brands/[brand-name]/seedance/[output-name]/.

---

Step 1 — Select brand

Scan ./brands/ for subfolders containing brand-identity/visual-guidelines.md.

• One brand: use it and confirm: "Using brand: [name]"
• Multiple: list and ask
• None: tell the user to run /brand first

---

Step 2 — Pick mode (T2V or I2V)

Ask:

"Text-to-video or image-to-video?

• Text-to-video — prompt only, Seedance generates the whole clip from scratch
• Image-to-video — start from one of your existing images and animate it"

If I2V, ask which image. Suggest:

brands/[brand-name]/product-shoot/
brands/[brand-name]/packshots/
brands/[brand-name]/styled/
brands/[brand-name]/clothing-shoot/
brands/[brand-name]/model-product-shoot/
brands/[brand-name]/ugc/

For an external upload:

mkdir -p brands/[brand-name]/seedance/uploads

"Drop your image into brands/[brand-name]/seedance/uploads/ and tell me the filename."

Wait for confirmation.

---

Step 3 — Prompt

Ask: "Describe the scene and motion you want. Be specific — what's happening, camera movement, mood, pace.

Examples:

• 'Slow dolly-in on a glass of iced espresso, condensation running down the side, warm afternoon light'
• 'A weathered hand picks up the jar, rotates it gently to show the label, sets it back down on the wood counter'
• 'Wide static shot, golden-hour wheat field, slight breeze moving the stalks, distant farmhouse in soft focus'"

If the prompt is bare ("rotate the product", "wind blowing"), suggest a refined version with camera language and atmosphere before firing.

---

Step 4 — Duration, resolution, aspect, tier, audio

Ask in one message:

• Duration — "How long? (3–12s, default 5s)"
• Resolution — "Resolution? 1080p (default), 720p, or 480p for cheaper drafts"
• Aspect ratio — for I2V, infer from the source image and ask only if ambiguous. For T2V, ask: 16:9 (default), 9:16, 1:1, 4:3, 3:4, 21:9
• Tier — "pro (default, highest quality) or lite (faster / cheaper)?"
• Audio — "Generate ambient audio? (off by default — adds cost. This is environmental sound, not voiceover.)"

---

Step 5 — Name the output

Derive a slug from the source image (if I2V) or the core idea of the prompt (if T2V):

• T2V: iced-espresso-pour, wheat-field-breeze, colette-kitchen-morning
• I2V: hey-bud-tint-rotate, jar-handpickup, meg-loaf-steam

Ask the user to confirm or rename.

---

Step 6 — Write seedance-spec.json

Create the output folder and write the spec:

brands/[brand-name]/seedance/[output-name]/
  seedance-spec.json

Text-to-video example:

{
  "output_name": "wheat-field-breeze",
  "brand": "[brand-name]",
  "mode": "text-to-video",
  "source_image": null,
  "prompt": "Wide static shot, golden-hour wheat field with slight breeze moving the stalks, distant farmhouse in soft focus.",
  "duration": 5,
  "resolution": "1080p",
  "aspect_ratio": "16:9",
  "tier": "pro",
  "generate_audio": false,
  "camera_fixed": true
}

Image-to-video example:

{
  "output_name": "hey-bud-tint-rotate",
  "brand": "[brand-name]",
  "mode": "image-to-video",
  "source_image": "brands/[brand-name]/product-shoot/hey-bud-tint-hand-hold/product-shot_v1.png",
  "prompt": "The product rotates 90 degrees clockwise, slow and smooth, light catching the label.",
  "duration": 5,
  "resolution": "1080p",
  "aspect_ratio": "9:16",
  "tier": "pro",
  "generate_audio": false,
  "camera_fixed": false
}

---

Step 7 — Generate

Run from the project root:

python3 skills/references/generate-seedance.py brands/[brand-name]/seedance/[output-name]

The script:

• Loads FAL_KEY from .env
• Routes to the right Seedance endpoint based on mode + tier + generate_audio
• Uploads the source image (I2V only)
• Runs the FAL job
• Downloads the output to [output-name]_v1.mp4, _v2.mp4, etc.

---

Step 8 — Present the result

Tell the user the exact path: brands/[brand-name]/seedance/[output-name]/[output-name]_v1.mp4

Ask: "Happy with this? Or regenerate, adjust the prompt, change tier, or extend duration?"

To regenerate (same spec):

python3 skills/references/generate-seedance.py brands/[brand-name]/seedance/[output-name]

To adjust: edit seedance-spec.json and re-run.

---

Notes

• Seedance Pro 1.0 supports up to 12s clips at 1080p. Lite tier is faster and cheaper but with a small quality drop.
• For I2V, Seedance respects the source image's composition closely — useful when you want motion without re-rolling the scene
• For T2V, Seedance is strong at multi-shot scene continuity and natural camera language
• camera_fixed: true locks the camera to a static position — useful for product spins, slow drifts, or when you want the subject (not the camera) to drive the motion
• Audio adds a separate generation pass; expect longer wait and added cost. Off by default.
• Each brand has its own seedance/ subfolder — multiple brands run side by side
• For Kling-based motion (with multi-image product element conditioning, start+end frame control), use /add-motion instead. Seedance and Kling are complementary, not redundant.