# Trend To Video Pack

> Trend-to-Video Pack

- Skill: `afterduang/trend-to-video-pack` (Agent Skill, multi-file: 3 files)
- Install (CLI): `npx skillmds@latest add afterduang/trend-to-video-pack`
- Raw SKILL.md: https://api.skillmd.com/api/skills/afterduang/trend-to-video-pack/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: AfterDuang (https://skillmd.com/u/afterduang)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/afterduang/trend-to-video-pack

---


# Trend-to-Video Pack

Convert a video idea into a practical handoff for video generation. Treat trend discovery as evidence gathering, not as a substitute for the user's creative judgment.

## Required brief

Collect or infer these fields before researching. Ask only for a field that materially changes the output; default to `TikTok`, `9:16`, `15 seconds`, and the user's stated country if available.

```text
idea, target platform, target country, time window, duration, aspect ratio,
non-negotiables (person/product/brand), desired use, preferred video model
```

State all assumptions in a compact `Assumptions` block.

## Workflow

### 1. Produce a trend-search plan

Translate the brief into 3-6 focused queries. Include a Creative Center plan when TikTok is the target:

```text
Country: <country>
Time: <time window>
Creative Center surfaces: Videos, Hashtags, Songs, Creators as relevant
Queries: <keyword, synonym, effect/format query>
Signals to compare: recency, engagement/growth if visible, format repetition,
and fit to the brief.
```

Use available web or browser tools to collect public candidate links. When a Creative Center page is available in the user's authenticated browser, read only visible public/authorized content and preserve the displayed country and time window. Do not bypass login, CAPTCHAs, paywalls, rate limits, or access controls.

If direct retrieval is unavailable, ask the user to provide 1-3 selected URLs, screenshots, or downloaded clips. Continue with the rest of the workflow; never invent live trend rankings, view counts, or facts about an unseen video.

### 2. Shortlist and let the user select

Present at most 10 candidates. For each, give the URL, observable source/date/region signal, and one sentence on why its format fits. Label unverified or web-indexed evidence clearly.

Do not analyze a video in detail until the user selects it and provides a playable/uploaded file, accessible URL, or video-analysis result.

### 3. Analyze the selected reference

Use a video-capable multimodal model when available. For an interactive Gemini workflow, give the user this prompt verbatim and request its structured result back:

```text
Analyze this reference video for producing a new video in the same content format.
Return JSON only with: duration, aspect_ratio, hook, pacing, shots[{start,end,
framing,camera_motion,subject,action,expression,wardrobe,scene,lighting,
transition}], audio_or_beat_notes, on_screen_text, reusable_format_elements,
asset_requirements, continuity_rules, and risks_or_uncertainties.
Describe observed actions precisely but do not identify private people or infer
sensitive traits. Separate observable facts from uncertain inferences.
```

For choreography, sport, or precise hand actions, recommend pose/hand tracking in addition to a general video model. General video understanding is sufficient for story, style, scene, product, and camera analysis, but is not a reliable substitute for frame-accurate motion data.

### 4. Identify the asset gap

Return assets in exactly three buckets:

```text
Required from user: likeness-critical or brand-critical material
Optional from user: material that increases likeness/control
Safe for AI to create: generic background, lighting, props, effects, and style variants
```

For a person or recurring character, request a front, side, and full-body reference plus wardrobe reference. For a product, request legible front/side/detail images and brand constraints. Do not ask for material that will not affect controllability.

### 5. Assemble the production pack

Read `references/production-pack-schema.md` and return its sections in that order.

Write clip prompts rather than one oversized prompt when the video changes location, framing, camera motion, subject action, or story beat. Group continuous action in one setting into a single clip when that improves motion continuity. Use 2-5 second clips by default; use a longer clip only for an intentional one-take.

Repeat the global continuity anchor in every clip prompt. When the generation tool supports start or end frames, recommend passing the prior accepted clip's last frame into the next clip. Otherwise, export that frame and use it as a reference image.

### 6. Preserve provenance and limits

Treat selected trends as format references. Avoid copying recognizable people, unlicensed media, watermarks, or an exact creator's protected visual identity. For cover/remake requests, label the output as a cover/remake plan and retain the source URL, creator attribution if visible, and music/source notes for the user to review before publishing.

## Output quality bar

Make every item actionable. A production pack is complete only if a producer can answer:

1. What are we making and why does it fit the selected trend?
2. Which assets must be supplied before generating?
3. What exact clip should be generated next?
4. What must remain consistent across clips?
5. What needs manual review before posting?