jackming568-commits/brand-promo-video-director-skill

Brand Promo Video Director

Operating Principle

Treat the video model as a cinematography and motion tool, not as the final compositor. Keep brand-critical elements such as logos, Chinese text, QR codes, booth numbers, product UI, data figures, certificates, legal labels, and subtitles stable through post-production overlays whenever possible.

The target is not "more motion"; the target is premium brand confidence: stable camera movement, clean hierarchy, readable information, cinematic energy, and audio that makes the value proposition easy to hear.

Workflow

1. Audit source assets before writing prompts. Check resolution, compression, text readability, logo clarity, QR codes, certificate details, and whether the asset can survive motion.
2. Define the distribution context: official video account, trade show screen, customer/private-domain sharing, social feed, sales follow-up, or internal pitch.
3. Build a scene map. Assign each scene a job: hook, capability, product proof, data proof, trust proof, case proof, company proof, call to action.
4. Decide what AI video generation should animate and what post-production should preserve. Never let the model invent certificates, UI text, QR codes, numbers, booth details, or brand names.
5. Write prompts with controlled motion. Prefer dolly, parallax, light sweeps, UI reveals, data flow, floating glass layers, and slow cinematic camera moves over aggressive shaking or over-fast transitions.
6. Generate one scene at a time and review before continuing when quality risk is high.
7. Composite clean text, logo, QR code, subtitles, data, and certificates after generation.
8. Mix audio after the visual edit. Narration must remain intelligible above the music. Use ducking under voiceover and avoid harsh or synthetic voice delivery.
9. Score the result with the rubric in references/score-rubric.md.
10. If the result is below target, diagnose whether the issue is source quality, AI generation, prompt design, scene timing, post-production, or audio mix before regenerating.

Asset Audit

Use references/source-asset-audit.md when evaluating whether images are good enough for a high-end video. A 1920x1080 image can still be insufficient if it has compressed text, small UI, fake-looking certificate detail, or too much information for motion.

Quality guidance:

• 1080p source is acceptable for 1080p delivery only if key text is sharp.
• 2K/4K source or vector/layered source is preferred for premium delivery.
• Small Chinese text, QR codes, and certificates should be rebuilt as overlays.
• Product screenshots and phone UI should be stabilized or recreated as flat layers if the model distorts them.
• For 8.5+ quality, prefer a layered workflow over pure image-to-video generation.

Prompting

Use references/prompt-templates.md for reusable AI video prompts, negative prompts, and scene-specific prompt blocks.

Prompt rules:

• State the intended camera move and speed.
• Ask for premium corporate technology style, not generic sci-fi clutter.
• Keep motion smooth, slow enough to read, and cinematic.
• Preserve layout and composition from the reference image.
• Forbid text mutation, logo mutation, fake documents, fake medals, and unstable shaking.
• For official B2B videos, use controlled energy rather than chaotic trailer motion.

Post-Production

Use references/post-production-guidelines.md for compositing, audio, subtitles, export, and quality control.

Post-production priorities:

• Re-overlay exact brand text and important Chinese copy.
• Re-overlay QR code and contact information from clean source.
• Rebuild data cards, titles, and lower thirds when AI output is soft.
• Use music that supports the message, but keep voiceover dominant.
• Use professional narration when the video is for public official channels.
• Export high-bitrate 1080p or create a higher-resolution intermediate before final delivery.

Review Behavior

When reviewing a draft, lead with what prevents the video from reaching the target quality. Be direct about blur, unstable motion, fake text, fake awards, weak audio, poor music fit, pacing problems, and brand trust issues.

When the user wants to continue generating scenes, avoid repeating failed patterns. Explain the next scene's intent, motion design, preservation requirements, and quality risks before generating if the previous draft had issues.

For serious brand work, prefer fewer high-quality scenes over more noisy scenes.