# Reusable Prompts

## Product Interview

Answer briefly:

1. Product URL or product name.
2. Who buys it?
3. What pain does it solve?
4. What is the one-sentence promise?
5. What objection stops people?
6. What is the offer and CTA?
7. Which platform matters first?
8. What is your account baseline there?
9. Do you have product photos, app recordings, creator photos, or saved characters?

## Reference Intake

Paste 3 to 10 TikTok or IG references.
For each, include any metrics you have:

- URL.
- Views.
- Account's usual views.
- Caption or transcript if available.
- Why you think it worked.

If you have Apify, paste the output here.
If not, plain links and notes are enough.

## Outlier Score

Make a table:

`label | URL | views | baseline | outlier score | hook | open question | sequence | payoff | risk | keep?`

Use `views / baseline` when baseline exists.
If baseline is missing, mark the score as `clear`, `likely`, or `unproven`.
Pick 3 to 5 winners.

## Reverse Engineering

For each selected reference, answer:

- What is the hook?
- What question keeps the viewer watching?
- What happens in each beat?
- Where is the proof?
- What is the payoff?
- What sound and setting make it feel real?
- What must we avoid copying?
- What inspired-by version fits this product?

## Test Matrix

Create 6 to 10 variants.
Change one variable per row.

Columns:

`variant_id | reference | variable changed | hook | creator | starting frame | proof | CTA | agent-media input notes`

Pick one first test take.
Explain why it is the safest first render.

## Starting Frame Brief

Write a first-frame brief:

- Real phone framing.
- Creator visible and believable.
- Product or app context visible only if useful.
- No text on face.
- Normal skin detail.
- Natural hands.
- Scene sound named.
- One clear action.

## Agent-Media Test Take Brief

Write the exact tool input for one test take.

If using `make_ugc`, include only:

- `script`
- one of `person`, `image`, or `character`
- optional `scene`
- optional `product_image`
- optional `duration`
- optional `captions`
- optional `caption_style`
- `look`
- `aspect_ratio`

If using `generate_video`, include only:

- `prompt`
- `seconds`
- `aspect`
- optional `model`
- optional `first_frame`
- optional `refs`
- optional `video_refs`
- optional `audio_refs`
- optional `quality`
- `audio`
- `request_id`

Ask for approval before rendering if this is not clearly the free first ad.

## Quality Review

Review the output:

- Lips.
- Hands.
- Face drift.
- Product clarity.
- Sound believability.
- Starting frame.
- Hook clarity.
- Payoff.
- CTA.

Verdict:

- `ship`
- `fix then batch`
- `remake first`

Give the shortest useful fix list.

## FFmpeg Finish

Use real app screen recording.
Make 9:16 MP4.
Add captions from the final transcript.
Keep face and captions title-safe.
Name files by variant ID.

If a file path is missing, ask for the recording or skip that finish step.

## Publish Packet

Create:

- File or URL.
- Variant ID.
- Platform.
- Caption.
- CTA.
- UTM link.
- Thumbnail note.
- Optional Postiz note.
- Baseline metric to compare against.

Judge each platform separately after posting.