{
  "skill_name": "video",
  "evals": [
    {
      "id": 1,
      "prompt": "We need a 2-minute product demo video for our SaaS homepage. What's the fastest way to produce it?",
      "expected_output": "Should check for product-marketing.md first. Should walk through the Product Demo Video workflow: script the key features and value props (cross-reference copywriting skill), screen record the product flow, programmatic overlay with Hyperframes or Remotion for titles/callouts/transitions, optional AI B-roll with Veo/Runway for establishing shots, voiceover via recording or AI avatar (HeyGen) for narration, export at platform-appropriate specs (16:9 for homepage). Should recommend Hyperframes for agent-friendliness (plain HTML, no React DSL). Should remind: don't use AI for product UI screens (models hallucinate UI) — use real screen recording. Should mention captions are essential (85% of social video watched without sound — applies to homepage too).",
      "assertions": [
        "Checks for product-marketing.md",
        "Walks through Product Demo workflow steps",
        "Uses real screen recording, not AI generated UI",
        "Recommends programmatic overlay tool",
        "Mentions captions",
        "Cross-references copywriting skill"
      ],
      "files": []
    },
    {
      "id": 2,
      "prompt": "We want to make weekly product update videos. About 60 seconds each. Don't want to be on camera. Recommend a setup.",
      "expected_output": "Should recommend an AI avatar workflow given recurring weekly cadence and no-camera preference. Should recommend HeyGen specifically: best lip-sync, has an MCP server (so agents can generate videos directly), 230+ avatars, 140+ languages, Creator plan supports unlimited 5-minute videos. Should explain custom avatars (upload 2-5 min of yourself for a digital twin) as an option for brand consistency. Should outline the recurring pipeline: script written from product context, HeyGen generates avatar video, optional programmatic overlay with Hyperframes for UI screenshots/callouts, export and distribute. Should mention this is exactly the case where AI avatars shine vs other approaches (recurring content, multilingual versions, personalized outreach at scale). Should warn: if authentic founder content matters more than scale, film yourself instead.",
      "assertions": [
        "Recommends AI avatar approach",
        "Names HeyGen specifically",
        "Mentions HeyGen MCP server for agents",
        "Mentions custom avatars option",
        "Identifies as a recurring use case",
        "Warns about authenticity tradeoff"
      ],
      "files": []
    },
    {
      "id": 3,
      "prompt": "I want to generate a 10-second clip of a person typing on a laptop in a coffee shop for our landing page. Which AI tool?",
      "expected_output": "Should apply the AI Video Generation model comparison. Should recommend Veo 3 for highest quality with synced audio, Runway Gen-4 for motion control and temporal consistency (~10 sec/gen sweet spot), or Kling 3.0 for lower-cost volume production. Should give a structured video prompt example following Subject + Action + Camera + Style + Mood pattern: 'A close-up shot of hands typing on a laptop keyboard in a cozy coffee shop, shallow depth of field, warm afternoon lighting through a window, camera holds steady, cinematic color grading, 4K.' Should warn about common mistakes: too vague, ignoring camera movement, forgetting style, requesting readable text. Should mention Sora has had limited availability — check current status.",
      "assertions": [
        "Compares Veo, Runway, and Kling",
        "Provides structured video prompt example",
        "Follows Subject + Action + Camera + Style + Mood pattern",
        "Warns about common prompt mistakes",
        "Notes Sora reliability caveats"
      ],
      "files": []
    },
    {
      "id": 4,
      "prompt": "We just did a 60-minute webinar. How do we get short clips out of it for social?",
      "expected_output": "Should apply the Repurposing Workflow: long-form content → Descript (clean up, remove filler, polish) → Opus Clip (auto-extract 5-10 best moments, scores virality potential) → CapCut (add captions, effects, platform styling) → distribute to TikTok, Reels, Shorts, LinkedIn. Should explain when to use each tool: Descript for transcript-based editing, Opus Clip for finding the best moments at scale, CapCut for platform-native polish, Captions.ai for auto-captions and eye-contact correction if needed. Should mention 85% of social video is watched without sound — captions are essential. Should mention aspect ratio matters: 9:16 for TikTok/Reels/Shorts, 1:1 or 9:16 for LinkedIn. Should recommend hooking in the first 3 seconds — cross-reference social skill.",
      "assertions": [
        "Applies repurposing workflow",
        "Names Descript, Opus Clip, CapCut in sequence",
        "Mentions captions essential",
        "Specifies aspect ratios per platform",
        "Mentions hooking in first 3 seconds",
        "May cross-reference social skill"
      ],
      "files": []
    },
    {
      "id": 5,
      "prompt": "We need to generate 50 personalized intro videos for sales outreach. Each one mentions a different company name and pain point.",
      "expected_output": "Should recommend an agent-native pipeline combining HeyGen MCP (or API) for the avatar narration + Hyperframes for any visual overlays. Should explain: prepare a master script template with variables, run a loop generating 50 HeyGen videos each with a personalized script, optional programmatic overlays via Hyperframes for company logo or visual context. Should note HeyGen is well-suited to personalized outreach at scale and has an MCP server. Should warn about quality tradeoffs at volume and recommend testing the first 5 manually before generating all 50. Should mention reply tracking to measure ROI vs cold text emails — these are expensive to produce so should outperform email significantly to justify the effort. Should mention captions for the videos.",
      "assertions": [
        "Recommends HeyGen + Hyperframes pipeline",
        "Names HeyGen MCP server",
        "Suggests template + loop approach",
        "Recommends testing 5 manually first",
        "Mentions reply tracking / ROI",
        "Mentions captions"
      ],
      "files": []
    },
    {
      "id": 6,
      "prompt": "Should I use Hyperframes or Remotion for programmatic video?",
      "expected_output": "Should compare the two based on the When to Pick Which table. Should recommend Hyperframes if: agent-driven (plain HTML/CSS, no React DSL — AI models generate better HTML than React components), minimal learning curve, basic animation needs, local rendering is fine, want Apache 2.0 license. Should recommend Remotion if: already a React shop, need complex animations (Spring, interpolate), need large-scale batch rendering via Lambda for AWS scale, can handle the React + Remotion API learning curve, comfortable with the company license for commercial use. Should note Hyperframes is from HeyGen and LLM-native by design. Should ask about the user's tech stack and animation complexity to recommend a final choice.",
      "assertions": [
        "Compares the two with the When to Pick Which table",
        "Notes Hyperframes uses plain HTML/CSS",
        "Notes Remotion supports Lambda for scale",
        "Mentions Apache 2.0 vs company license",
        "Recommends Hyperframes for agent-driven workflows",
        "Asks about stack or animation needs"
      ],
      "files": []
    },
    {
      "id": 7,
      "prompt": "There's a TikTok edit style I love — fast cuts, one-word captions that pop, a whoosh on every scene change. I have my own talking-head clip. Break down how that edit works so I can replicate the style. Here's the reference: [link]",
      "expected_output": "Should apply references/edit-anatomy.md (reverse-engineer the edit into a reusable spec), not just describe it. Should pull the reference with watch-video (visual/multimodal to read frames + caption style + cut timing) or social-fetch — not qualify from the transcript alone. Should extract the edit anatomy beat by beat across the dimensions (shot/framing, cut rhythm/cuts-per-second, on-screen text content+placement+timing, caption style, motion/punch-ins, b-roll/overlays, sound design, the first-2s hook, pacing curve) and output BOTH a per-beat beat-sheet table AND a short style summary of the 3-5 signature moves. Should emphasize patterns over instance-logging. Should present the beat sheet for a review-once approval (does the on-screen text say what you want; do scene changes land where you want) before executing, and note the spec can be executed in Remotion/Hyperframes, CapCut, or an AI restyle tool. Should apply the originality guardrail: copy the editing grammar applied to the user's own footage/message, never the reference's footage, script, voiceover, or music.",
      "assertions": [
        "Applies the edit-anatomy reverse-engineering method, not a plain description",
        "Pulls the reference with watch-video/social-fetch to read the actual frames, not just the transcript",
        "Extracts the edit anatomy across the dimensions and expresses patterns (not a raw list of cut timestamps)",
        "Outputs a per-beat beat sheet AND a style summary of the signature moves",
        "Presents the beat sheet for a review-once approval before executing",
        "Notes execution paths (Remotion/Hyperframes, CapCut, or AI restyle tool)",
        "Applies the originality guardrail — copies editing grammar applied to the user's own footage, never the reference's footage/script/music"
      ],
      "files": []
    }
  ]
}

# AI Video Prompting Guide

How to write effective prompts for AI video generation models (Veo, Runway, Kling, Pika).

---

## Prompt Structure

A strong video prompt follows this formula:

```
[Subject] + [Action] + [Camera movement] + [Visual style] + [Lighting/mood] + [Technical specs]
```

### Example Prompts by Use Case

**Product hero shot:**
```
A sleek laptop on a minimal white desk, screen glowing with a dashboard UI,
camera slowly orbits 180 degrees around the desk,
soft volumetric lighting from the left, shallow depth of field,
cinematic commercial aesthetic, 4K
```

**Lifestyle B-roll:**
```
A woman in a modern co-working space smiling while looking at her phone,
natural window light, candid documentary feel,
camera handheld with subtle movement, warm color grading
```

**Abstract/brand:**
```
Flowing liquid gold particles forming the shape of a network graph,
dark background, particles catch light as they move,
slow-motion macro photography style, dramatic rim lighting
```

**SaaS explainer scene:**
```
An overhead shot of a team around a conference table pointing at charts,
camera slowly pushes in, bright modern office,
clean corporate style, even lighting, 1080p
```

---

## Camera Movement Vocabulary

Use these terms — video models understand them:

| Term | Effect |
|------|--------|
| **Static** | Locked camera, no movement |
| **Pan left/right** | Camera rotates horizontally |
| **Tilt up/down** | Camera rotates vertically |
| **Dolly in/out** | Camera moves toward/away from subject |
| **Orbit** | Camera circles around subject |
| **Tracking shot** | Camera follows moving subject |
| **Crane/aerial** | Camera rises or descends |
| **Handheld** | Subtle shake, documentary feel |
| **Zoom** | Lens zoom (different from dolly) |
| **Slow push** | Gradual dolly in — builds tension/focus |

---

## Style Keywords

### Cinematic
- "cinematic color grading"
- "anamorphic lens flare"
- "shallow depth of field"
- "film grain"
- "35mm film"

### Commercial/Corporate
- "clean commercial lighting"
- "bright and airy"
- "professional corporate aesthetic"
- "even, diffused lighting"

### Documentary
- "handheld documentary style"
- "natural lighting"
- "candid, unposed"
- "observational camera"

### Social/Trendy
- "vertical 9:16"
- "fast-paced cuts"
- "bold text overlays"
- "high contrast, saturated colors"

---

## Model-Specific Tips

### Veo (Google)

- Excels at photorealism and complex scenes
- Supports audio generation synced to video
- Best with detailed, descriptive prompts
- Specify "high resolution" or "1080p" for best quality
- Can handle multiple subjects and scene transitions

### Runway Gen-4

- Strong motion control — specify camera movements precisely
- Best temporal consistency (subjects stay consistent across frames)
- Use motion brush for specific area animation
- Image-to-video works well — provide a reference frame
- Keep prompts under 100 words for best results

### Kling

- Can generate up to 2 minutes (much longer than others)
- Good for longer narrative sequences
- More affordable for bulk generation
- Quality drops slightly at longer durations
- Best with simpler scenes and fewer subjects

### Pika

- Fastest generation time (under 2 minutes)
- Good for quick iterations and experimentation
- Effects mode adds motion to still images
- Best for short clips (5-15 seconds)
- Less control over camera movement

---

## Common Prompt Mistakes

| Mistake | Why It Fails | Fix |
|---------|-------------|-----|
| "A person using our app" | Too vague, no visual detail | Describe the person, setting, lighting, camera |
| Including text/logos | AI can't render readable text | Add text in post via Hyperframes/CapCut |
| "Make it viral" | Not a visual instruction | Describe the visual style you want |
| Extremely long prompts (200+ words) | Models lose focus | Keep to 50-100 words, be specific |
| No camera direction | Random/static camera | Always specify movement or "static" |
| "Realistic" alone | Not specific enough | "Photorealistic, natural lighting, shot on RED camera" |

---

## Prompting Workflow

1. **Reference first** — find a real video that looks like what you want
2. **Describe it** — break down: subject, action, camera, style, mood
3. **Generate 3-4 variations** — same concept, different angles or styles
4. **Iterate on the best** — refine the prompt based on results
5. **Composite** — combine AI footage with programmatic text/overlays

---

## Aspect Ratios

Always specify in your prompt or generation settings:

| Platform | Ratio | Resolution |
|----------|-------|-----------|
| YouTube | 16:9 | 1920x1080 or 3840x2160 |
| TikTok/Reels/Shorts | 9:16 | 1080x1920 |
| Instagram Feed | 1:1 or 4:5 | 1080x1080 or 1080x1350 |
| Website hero | 16:9 | 1920x1080 |
| LinkedIn | 16:9 or 1:1 | 1920x1080 |

---

## Cost Optimization

- **Iterate at low resolution** — upscale only the final version
- **Use Kling for drafts** — cheapest per second, switch to Veo/Runway for finals
- **Image-to-video** — providing a reference frame saves generation credits and gives better results
- **Batch similar prompts** — models often offer volume discounts
- **Cache and reuse** — B-roll clips can be reused across multiple videos

# Reverse-Engineering an Edit (The Beat Sheet)

A viral short-form video usually isn't winning on the footage — it's winning on the *edit*: the cut rhythm, the caption style, the punch-ins, the on-screen text landing on the exact word, the b-roll cutaways, the sound design. This reference turns a reference edit you admire into a **reusable edit spec** — a beat sheet you (or an editing tool) can execute against your own footage — without copying a single frame of theirs.

This is the tool-agnostic half of "copy any viral edit": the *decomposition*. The generation is whatever you edit with afterward — CapCut, Premiere, Remotion/Hyperframes, or an AI restyle tool. The spec is the deliverable.

## When to use it

- A competitor's or creator's edit keeps stopping your scroll and you want to understand *why* and replicate the technique
- You have raw footage (a talking-head clip, a demo) and a reference edit whose style you want to match
- You're briefing an editor or a template and need the edit decisions written down, not vibes

Don't use it to copy someone's actual creative — this extracts the *editing grammar* (structure, rhythm, caption treatment), not the script, footage, or brand. Same rule as mining organic content for vocabulary in the hook system: take the technique, never the creative.

## Step 1 — Pull the reference so you can actually read the edit

You cannot decompose an edit from a description of it. Get the frames and the timing:

- **watch-video** (visual or multimodal mode) — extracts the transcript *and* samples frames at the cut points, so you can read on-screen text, caption style, and shot changes. This is the primary tool.
- **social-fetch** — pull the post for the caption, engagement, and the media URL when the reference is a specific tweet/Reel/TikTok.
- Screenshots of key frames also work if the user supplies them — you need the visual, not just the words.

Note the total duration and roughly how many cuts there are before you start — cuts-per-second is the single most telling number about an edit's energy.

## Step 2 — Extract the anatomy, beat by beat

Walk the reference from 0:00 and log every editing decision. The dimensions that define a short-form edit:

| Dimension | What to read off the reference |
|---|---|
| **Shot & framing** | Talking head / screen recording / b-roll / text card; close-up vs. wide; headroom, rule-of-thirds, or dead-center |
| **Cut rhythm** | Where each cut lands and how fast (cuts-per-second); is it on the beat, on the word, or on the breath? |
| **On-screen text** | The words, when each appears/disappears, and *where* on the frame (top-third caption vs. big centered statement) |
| **Caption style** | Font, weight, color, outline/box, and animation (word-by-word pop, karaoke highlight, whole-line) |
| **Motion** | Punch-ins / zoom pushes, shakes, whip-transitions, speed ramps — where and how aggressive |
| **B-roll & overlays** | Cutaways, stickers, arrows, emoji, screenshots, meme inserts — what's laid over the base footage and when |
| **Sound design** | Music choice and where it hits, SFX (whooshes, dings, risers), and deliberate silence before a beat |
| **Hook (first 2s)** | The single most-copied element — what's on screen and said in the opening two seconds, before anyone's committed |
| **Pacing curve** | Does it stay frantic, or fast-hook → slower-body → fast-CTA? Map the energy over the runtime |

Read the *pattern*, not just the instances: "a hard cut + punch-in on every new sentence," "caption is one word at a time, yellow, karaoke-highlighted, bottom third," "a whoosh SFX on every scene change." Patterns are what make an edit replicable; a list of 40 individual cuts is not.

## Step 3 — Write the beat sheet

Two artifacts: a per-beat table and a short style summary.

**The beat sheet** — one row per beat (a beat = a cut or a distinct edit event):

```
| Beat | Time      | Shot            | On-screen text        | Caption style        | Transition / motion   | Audio            |
|------|-----------|-----------------|-----------------------|----------------------|-----------------------|------------------|
| 1    | 0:00–0:02 | CU talking head | "STOP doing this"     | word-pop, yellow, ctr| hard in, slow push    | music in + riser |
| 2    | 0:02–0:04 | screen record   | (caption only)        | karaoke, white, btm  | hard cut + whoosh     | click SFX        |
| …    |           |                 |                       |                      |                       |                  |
```

**The style summary** — the 3–5 *signature moves* that make this edit recognizable, stated so they're reusable:
- e.g. "Every sentence gets a hard cut + a 5% punch-in." / "Captions are one word at a time, bottom-third, karaoke-highlighted." / "A whoosh SFX on every cut; music drops out for 0.5s before the CTA." / "The hook is a bold centered statement on frame 1, no logo."

The signature moves are the real deliverable — someone can apply those five rules to any footage and get the style. The table is the detailed backup.

## Step 4 — Review once, then execute

Show the beat sheet before anyone edits anything — the same review-once gate as the ad-creative creative review page. The reviewer checks two things:

- **The on-screen text says what you want** (mapped to your message, not the reference's)
- **The scene changes land where you want them** (your footage's beats, not a blind copy of the reference's timing)

Approve, then execute the spec with your footage:
- **Remotion / Hyperframes** — when you want the edit templated and data-driven (see the programmatic-video section in SKILL.md); the beat sheet *is* the composition spec.
- **CapCut / Premiere / an editor** — hand off the beat sheet + style summary as the brief.
- **An AI restyle tool** — feed the style summary as the target style.

## Originality guardrail

You are copying the *edit*, not the content. The beat sheet describes technique (cut rhythm, caption treatment, motion, sound design) applied to **your** footage and **your** message. General editing techniques and style cues are usually reusable — U.S. copyright protects expression, not procedures or methods (17 U.S.C. §102(b)) — but the reference's specific creative expression is not, and closely reproducing a finished video's exact selection and arrangement of choices can still create risk. So copy the grammar, not the finished work: use your own footage, message, script, voiceover, licensed music/SFX/samples, and brand elements. If the reference's "style" is really a specific bit or sketch, that's their creative — draw inspiration, don't reproduce it.

## Common mistakes

- **Describing instead of reading** — you can't extract caption style or cut timing from the transcript alone; pull the frames (watch-video).
- **Logging instances, not patterns** — 40 cut timestamps isn't a spec; "hard cut + punch-in per sentence" is.
- **Copying the reference's timing onto different footage** — beats land on *your* words and *your* cuts; the reference gives you the grammar, not the calendar.
- **Skipping the hook** — the first 2 seconds carry most of the retention; decode them in the most detail.
- **Reproducing the creative** — matching the edit is fine; re-shooting their exact bit, script, or using their footage/music/SFX is not.