Video Production Strategy
Video production strategy and tool selection. Covers AI video generation model comparisons, programmatic video framework choices, video script structure design, and production workflow planning.
This skill is responsible for strategy planning and tool selection. The deployed
video-editingskill can perform actual ffmpeg editing operations (cropping, jump cuts, text overlays, speed changes, etc.).
Input
- Video goal (product demo / explainer / social media short / advertisement / tutorial)
- Target platform (Douyin / Bilibili / Kuaishou / WeChat Video / YouTube / Instagram)
- Existing assets (screenshots, screen recordings, scripts, brand materials)
- Budget and tech stack constraints
Output
- Recommended production plan (tool selection + rationale)
- Platform adaptation specs (resolution, duration, aspect ratio)
- Production workflow steps
- Video script structure (if needed)
Execution Steps
1. Gather Context
Confirm the following information (ask proactively if not provided):
- Video type: Product demo, explainer, social media short, advertisement, tutorial
- Target platform: Determines aspect ratio and duration limits (see platform specs table below)
- Presentation style: AI digital human / voiceover + visuals / fully programmatic / screen recording
- Asset availability: Whether existing assets exist (screenshots, screen recordings, logo)
- Reuse needs: One-time vs. template-based batch production
- Budget: Some tools charge per video duration
2. Select Production Plan
| Plan | Use Case | Tools | Availability |
|---|---|---|---|
| Programmatic video | Template-based, data-driven, batch | Hyperframes, Remotion | ✅ Node.js available |
| AI video generation | Original footage, B-roll | Veo 3, Sora 2, Runway, Kling, Seedance | ⚠️ Requires API key |
| AI digital human | Human presenter, multilingual | HeyGen, Synthesia | ⚠️ Requires API key |
| Editing/remix | Long video to short clips | ffmpeg (video-editing skill), CapCut | ✅ ffmpeg available |
3. AI Video Generation Model Comparison
If the user needs AI-generated video footage, refer to this table for selection:
| Model | Resolution | Max Duration | Specialty | Cost |
|---|---|---|---|---|
| Veo 3 (Google) | Up to 1080p | Variable | Best quality + synchronized audio | API billing |
| Sora 2 (OpenAI) | Up to 1080p | ~20s | Cinematic quality + synchronized audio | API + ChatGPT |
| Runway Gen-4 | Up to 4K | ~10s/clip | Motion control, temporal consistency | Subscription |
| Kling 2.5/3.0 (Kuaishou) | Up to 1080p | Up to 2 min | Long shots, low unit cost | Per second (low unit price) |
| Seedance (ByteDance) | Up to 1080p | Short clips | Fast generation, high motion fidelity, good for batch | Points-based |
| Hailuo / MiniMax (MiniMax) | Up to 1080p | Short clips | Cross-shot character consistency | Points-based |
| Pika 2.x | 1080p | Short clips | Fast effects, image-to-video | Points-based |
| Hunyuan Video / Wan 2 (Tencent) | 720p-1080p | Variable | Open-source self-deployment, fully controllable, no API fee | Free (requires GPU) |
Quick Selection Guide:
- Best quality + audio: Veo 3 or Sora 2
- Batch / low cost: Kling, Seedance
- Cross-shot character consistency: Hailuo
- Self-deployment / brand control: Hunyuan Video or Wan 2 (open-source weights)
- Image-to-video: Kling, Pika, Runway
The model version numbers and pricing in the table above iterate rapidly with vendors, for reference only. Always refer to the latest official vendor announcements.
For AI video prompting guidelines, see references/ai-video-prompting.md
4. Target Platform Specifications
Chinese Platforms
| Platform | Aspect Ratio | Recommended Resolution | Duration Limit | Notes |
|---|---|---|---|---|
| Douyin | 9:16 | 1080x1920 | 15s / 60s / 15min | Completion rate weight highest under 15s |
| Bilibili | 16:9 | 1920x1080 | No hard limit | Primarily landscape, portrait also supported |
| Kuaishou | 9:16 | 1080x1920 | 11min | Primarily portrait |
| WeChat Video | 9:16 / 16:9 | 1080x1920 | 30min | Short videos < 1min prioritized for recommendation |
| Xiaohongshu | 9:16 / 3:4 | 1080x1920 | 15min | 3:4 also commonly used |
International Platforms
| Platform | Aspect Ratio | Recommended Resolution | Duration Limit |
|---|---|---|---|
| YouTube | 16:9 | 1920x1080 / 3840x2160 | Unlimited |
| YouTube Shorts | 9:16 | 1080x1920 | 60s |
| TikTok | 9:16 | 1080x1920 | 10min |
| Instagram Reels | 9:16 | 1080x1920 | 90s |
| Instagram Feed | 1:1 / 4:5 | 1080x1080 / 1080x1350 | 60s |
5. Programmatic Video Frameworks
Generate videos with code, suitable for repeatable, template-based, data-driven batch production.
Hyperframes (HTML/CSS, recommended) – Open source, Apache 2.0, each frame is an HTML document, most agent-friendly. Remotion (React) – Stronger animation capabilities, suitable for complex motion effects and Lambda batch rendering.
| Factor | Hyperframes | Remotion |
|---|---|---|
| Agent friendliness | Better (pure HTML) | Good (React) |
| Animation capability | Basic (CSS transition) | Advanced (Spring, interpolate) |
| Batch rendering | Local | Lambda (AWS) |
| License | Apache 2.0 | Commercial license required |
6. Production Workflow Templates
Product Demo Video: Script → Screen recording → Programmatic overlays (titles, annotations) → AI B-roll → Voiceover → Export per platform specs
Explainer Video: Script (problem → solution → CTA) → Choose presentation style (AI digital human / voiceover) → Visual production → Add subtitles → Export
Batch Social Media Shorts: Build template in Hyperframes/Remotion → Inject data → Batch render → Add platform subtitles → Distribute
Viral Remix Replication: Pull reference video → Break down editing beat by beat → Output beat sheet → Execute with own footage. See references/edit-anatomy.md
7. Agent-Controllable Pipeline
Agent writes script (using Profile)
↓
Hyperframes: HTML → MP4 (programmatic video)
and/or
AI video API: Generate B-roll
↓
ffmpeg (video-editing skill): Edit and composite
↓
Outputs → outputs/
Common Mistakes
- Choosing tools before strategy – Clarify video goals first, then select tools
- Adding text inside AI-generated video – Models cannot reliably render readable text; use programmatic overlays
- Ignoring subtitles – 85% of social media videos are watched on mute; must include subtitles
- Wrong aspect ratio – Portrait 9:16 for social media, landscape 16:9 for YouTube/websites
- Overproduction – Authenticity often outweighs polish on social media platforms
Profile Awareness
Detects === EASEL ACCOUNT PROFILE === marker.
When Profile is present:
- Read brand visual style (colors, fonts, logo) from Profile and apply to video templates
- Read target platform preferences, prioritize corresponding platform specs
- Read audience persona, adjust video style (professional / casual / trendy)
- Read product information and use directly for script writing
When Profile is absent:
- Fall back to general mode, proactively ask about brand style and target platform
- Recommend plan based on general best practices