# Wouldliker Video Template Packs

## Quick Answer

Wouldliker video template packs are reusable prompt and shot-pattern recipes for AI-video agents. They sit on top of sound routes and `sound_behavior`.

They answer a practical question:

`I know the route. What should the video look like?`

The v1 template packs cover:

- Vlog: fashion recaps, routines, food lifestyle, travel days.
- Product Reveal: ecommerce reveals, sports card reveals, electronics feature reveals, food/package reveals.
- Presentation: tutorials, food reviews, shopping reviews, app demos.

## What One Template Contains

Each template includes:

- `template_id`
- `route`
- `sound_slug`
- `duration_s`
- `inputs`
- `prompt_template`
- `shot_pattern`
- `caption_template`
- `cover_frame_guidance`
- `selection_keywords`
- `sound_behavior_role`
- `payoff_required`

## Example

```json
{
  "template_id": "product_reveal_sports_card_30s",
  "route": "Product Reveal",
  "sound_slug": "product-reveal",
  "duration_s": 30,
  "inputs": ["card_name", "pack_name", "sport", "reaction_line"],
  "prompt_template": "Create a 30s TikTok sports trading card reveal...",
  "payoff_required": true
}
```

## How Agents Use It

An agent can call `/api/v1/clip-pack` with a topic. Wouldliker returns the sound route and a matching `template_pack`.

The agent can then replace placeholders such as:

- `{brand_name}`
- `{product_name}`
- `{card_name}`
- `{dish_name}`
- `{feature_name}`
- `{creator_name}`

This makes the response useful for faceless video pipelines, n8n workflows, Make, Zapier, ComfyUI, Remotion, ffmpeg services, and AI creative agents.

## Agent Prompt Pack

`/api/v1/clip-pack` also returns `agent_prompt_pack`. This is the stable prompt layer that a video agent or future LLM concierge can use directly.

It includes:

- `primary_prompt`: a filled, tool-agnostic prompt for AI-video generation.
- `short_prompt`: a compact prompt for workflow builders.
- `negative_prompt`: boundaries to avoid fake claims, bad frames, and forced drops.
- `structured_prompt`: duration, platform, sound role, first-second guidance, shot pattern, hooks, caption, and risk rationale.
- `missing_inputs`: placeholders the caller did not provide, such as `brand_name` or `product_name`.
- `clarifying_questions`: short questions an LLM can ask before finalizing the prompt.
- `llm_guardrails`: rules for DeepSeek or another model so it does not invent rights, proof, or guaranteed results.

This means DeepSeek can later act as a conversational layer on top of Wouldliker data: ask what the user is making, call `clip-pack`, fill the prompt pack, and return a ready production plan.

## Boundary

Template packs are not downloadable audio. They are planning and prompt assets.

The audio boundary remains:

TikTok-native route; confirm final sound availability and usage inside TikTok before posting. Not cleared for paid ads, brand/commercial use, or off-platform reuse unless rights are confirmed.

## Machine-Readable Files

- [Video template packs JSON](/data/video-template-packs.json)
- [Video template packs schema](/data/schemas/video-template-packs.schema.json)
- [OpenAPI spec](/api/openapi.json)
- [Sound behavior profiles](/sound-behavior-profiles/)
- [Sound brief spec](/sound-brief-spec/)