Make a short video worth finishing.

Use this workflow when you are asked to research, write, storyboard, build, revise, or validate a short-form video. Follow the requested platform and duration. If none are supplied, target 15 seconds or less and keep the video under 20 seconds. Do not force a word-rate target: time an actual read and simplify rushed copy.

Before drafting, identify the following. If a critical fact, permission, capability, or destination is missing, mark it Unknown and write around it.

• Audience, named vertical when supported, and the one problem, curiosity, or useful result.
• Evidence for the problem and each specific claim, number, quote, or product behavior. Separate observed facts, exact owner wording, inference, hypothesis, and illustrative scenario.
• For adaptations: source URL and retrieval date, plus the abstract mechanism. Never copy wording, footage, sound, identity, distinctive expression, or claims.
• Platform, aspect ratio, duration, audio approach, available motion and B-roll assets, and intended CTA.
• Whether generated people or voices appear and what disclosure the platform and asset record require.

A like or save is evidence of the researcher’s interest, not proof of audience-wide demand. A view count measures attention; it does not prove leads or sales.

Capture the original source, date, audience, pain in the owner’s words when available, visible artifact, consequence, uncertainty, and source or usage rights. If the pain is not confirmed, label it as a hypothesis and state how to validate it.

Write one sentence describing what the viewer will understand or be able to do after watching. Keep the promise proportional to the evidence. Choose one problem and one useful next step. Do not lead with an AI tool unless the source evidence shows the tool itself is the viewer’s problem.

Choose a single dominant mechanism, such as an object reveal, visible before-and-after, useful-system demonstration, numbered checklist, simple workflow, branching decision, or human handoff. Adapt only the abstract mechanism. Use new wording, examples, design, and proof.

Write the exact opening hook, complete spoken narration or full on-screen transcript, on-screen headline and labels, CTA, and captions. For every beat, specify start and end time, viewer purpose, exact words, visual action, transition, sound, and source or asset note.

• Make the premise understandable in the first three seconds and the full story clear with sound off.
• For narration, count the spoken words and estimate duration using an actual read.
• For silent video, hold each text block long enough to read comfortably. Reduce copy before speeding delivery.
• Keep display text short and captions accurate.

Download the brief and storyboard template to capture exact timing and assets.

Use motion to clarify the idea. Specify composition, hierarchy, entry and exit, transitions, camera movement, UI or artifacts, captions, sound, and safe areas. Reuse assets only after confirming they are portable and approved. Use local fonts for offline rendering; do not depend on network font imports.

For HyperFrames projects, preserve required composition IDs, `data-*` timing attributes, `.clip` classes, registered GSAP timelines, and existing project conventions. Do not casually rename selectors or change timing semantics. Run the project’s canonical validation command.

Choose one action. If the CTA promises a download, DM, lead capture, comment automation, or email sequence, verify the asset and the entire destination and delivery path from a clean test account. Record the test.

If it is not verified, use an honest non-delivery CTA or label it Draft; destination unverified. Never describe an automation as active without a successful end-to-end test.

Use only video, music, fonts, logos, and images that are owned, licensed, public domain, or specifically approved for the intended use. Record the source and clearance. If footage rights are uncertain, build the visual in code or hold the clip. Label fictional interfaces and examples Illustrative example so they cannot be mistaken for customer evidence.

Never use private customer data in mockups. Do not make a generated person appear to be a real customer or expert. Disclose generated people or voices wherever the user, platform, or applicable rules require it, and record the disclosure in the asset notes.

Run the documented project check, such as `npm run check` for the relevant HyperFrames project. Fix structural errors and triage overflow warnings. Inspect the review render at full speed and on a phone-sized display. Check:

• First-frame clarity, sound-off comprehension, caption accuracy, and real-read duration.
• Phone readability, safe areas, contrast, crop, transition continuity, and motion timing.
• Music and SFX rights, narration intelligibility, audio levels, and disclosures.
• Factual claims against cited sources; examples against their labels; CTA against its test receipt.
• Actual dimensions, frame rate, duration, playback, and remaining issues.

A passing code check is not a visual review. A rendered file is not approval to publish.

Return the paths to editable source and local assets, source URLs and retrieval dates, license and clearance notes, actual validation command and result, render inspection evidence, unknowns, and reviewer. Keep publication approval separate from production approval.

• One specific, sourced viewer problem or clearly labeled hypothesis.
• Original expression; no copied script, footage, sound, distinctive assets, identity, or unsupported claims.
• Complete hook, transcript or narration, CTA, timecoded storyboard, motion, and sound directions.
• Timing checked by an actual read or comfortable silent reading.
• Understandable without sound and legible on a phone.
• Fictional UI and examples are labeled; no fake customer proof.
• Every media asset has documented rights for the intended use.
• Generated people and voices are disclosed when required.
• CTA destination and delivery are verified or clearly marked as draft.
• Project validation passed and the full rendered video received visual review.
• Publication approval remains a separate decision.