Video Storyboard Prompt and Generation

Turn a short video idea into a production-ready timeline prompt, review it with the user, and generate the approved clip through Fal.

Loadout capabilities

Use <integration-tag>direct:fal_tools</integration-tag> for video generation. The normal path uses direct:fal_tools:fal_text_to_video; use an image-to-video Action from the same Integration only when the user supplied a compatible reference image and its live schema accepts that input.

The Skill provides guidance only. Inspect the live Action schema and cost before execution, and call the Action separately through Loadout.

Inputs

Collect or infer:

• Core subject and action.
• Intended duration from 4 to 15 seconds.
• Aspect ratio: 21:9, 16:9, 4:3, 1:1, 3:4, or 9:16.
• Visual style, mood, location, time of day, and lighting.
• Desired camera language and pace.
• Any text that must appear on Integration · Fald materially change the result. Default to 4 seconds, 16:9, and a coherent cinematic treatment when the user has no preference.

Storyboard method

1. State the visual premise in one sentence.
2. Split the duration into two to five beats. Keep the timing total exact.
3. Give each beat one dominant purpose: establish, reveal, develop, climax, or resolve.
4. Specify framing and motion with concrete terms such as wide shot, close-up, push-in, tracking, orbit, overhead, or pull-back.
5. Describe visible motion, lighting change, and transition behavior. Avoid contradictory camera directions.
6. Preserve subject, wardrobe, palette, and environment continuity across beats.
7. Put sound, dialogue, or on-screen text in a separate final clause. Do not promise exact audio or typography unless the live model supports it.

For reusable structures and a final quality checklist, read storyboard-patterns.md.

Prompt format

Produce one concise prompt in this order:

[style, duration, aspect ratio, mood]

0-Xs: [shot, subject, action, lighting]
X-Ys: [shot, subject, action, transition]
...

Continuity: [identity, wardrobe, environment, palette]
Sound/text: [optional requirements]
Avoid: [only material failure modes]

Show the prompt and the chosen duration, resolution, and aspect ratio before generating. If the user asked only for a prompt, stop after delivery.

Generation

When generation is requested:

1. Inspect direct:fal_tools:fal_text_to_video and preflight when available.
2. Present any runtime price or confirmation request exactly as returned.
3. Execute once after approval. Do not retry if the result already contains an asset ID, video URL, or retryable: false.
4. Return the generated asset and the exact prompt used.

If the provider rejects the request, report the returned reason, revise only the incompatible part, and ask before another paid attempt.

Boundaries

• Do not claim a specific underlying video model unless the live Action identifies it.
• Do not claim perfect character identity, lip sync, readable text, or frame-exact camera control.
• Refuse disallowed sexual, exploitative, deceptive, or privacy-invasive content.
• Treat uploaded people and branded assets as user-provided material whose rights and consent must be confirmed when unclear.

Attribution

This is a native Loadout re-authoring informed by the storyboard method in the pinned source described in UPSTREAM.md. The preserved source license is in LICENSE.txt. It is not a byte-for-byte import.