AI Blueprint — Static Ad Production

Role: you are my head of creative strategy for an e-commerce brand selling on Meta.

Context
- My brand: [BRAND] — [PRODUCT] — [OUR WEBSITE URL]
- The competitor: [COMPETITOR] — [URL] — they sell the same product under a different brand (same product, same promise).
- I'm giving you a folder of 35 of their ads: [FOLDER] (static images + videos).
- I'm also giving you a folder of winning templates, sorted by concept: [TEMPLATES FOLDER].

Tools to use
- The ad-blueprint-gpt-image-2 skill (JSON + ChatGPT Images 2.0 prompts).
- mr-transcribe to transcribe the videos (OpenAI API).
- The TrendTrack MCP for competitive intelligence.

Mission, in this order
1. Full market research: analyze every static, transcribe every video, then complete it with TrendTrack on the competitor AND on our brand. Deliverable: top angles, top headlines, avatar, winning formats, offer, weaknesses (in the format of the provided template).
2. 1:1 adaptation of the winning templates: for each concept, adapt every ad to every winning angle with its headlines, double every creative, aggressive conversion-driven headlines. One JSON per creative (50 to 60 total).
3. Generation with GPT Image 2 (quality high, 1:1), then QC of every image by a dedicated agent.
4. Delivery: a comparison site (template vs our ad), then a demo link with only the final images.

Rules
- Nothing invented: every angle and every headline points back to its source ad.
- Headlines in English, quoted word for word in the research, then rewritten for our creatives.
- Get my sign-off on the market research before moving to step 2.

Here is the folder of 35 ads from [COMPETITOR]: [FOLDER].

1. Transcribe every video with mr-transcribe (API, never local whisper).
2. For EACH ad, fill one row: # · format (static/video) · exact headline (word for word) · subheadline · video hook (first 3 seconds) · angle · mechanism used · visual format · proof shown · offer · CTA.
3. Group the ads by angle. For each angle: the idea in 2 sentences, the number of ads, duplicates, the best headlines.
4. Rank the angles by weight (volume + duplicates + run time).

Don't summarize too early: give me the 35-ad table first, then the synthesis.

Using the TrendTrack MCP, pull as much information as possible on [COMPETITOR] AND on [BRAND]:
- Meta pages, number of active ads and trend over 6 months;
- scaling ads and most duplicated ads;
- available video transcripts;
- website traffic and 30-day trend;
- offers, prices, guarantees, sales pages;
- emails and promotions.
Cross-check with the 35-ad analysis: confirm or correct the angle ranking. Cite the source of every number.

Write the final market research report, in exactly this structure:

1. Top 5 angles — table: # · Angle · The idea (2-3 sentences, including the mechanism) · Weight.
2. 5 secondary angles — one line each.
3. Top 5 headlines — word for word.
4. Headline bank — sorted by category: Pain & mechanism · Comparison · Product · Story · Urgency.
5. Video hooks — word for word.
6. Avatar — age, situation, visible problem, what they've tried and spent, everyday shame behaviors.
7. What proves it to them — concrete everyday proof.
8. Winning formats — with the static / video split.
9. Their offer — price, subscription, guarantee, social proof.
10. Their weaknesses — backed by TrendTrack numbers.
11. Recommendation for [BRAND] — the 3 angles and 10 headlines to attack first.

These angles and headlines become the source for our future creative JSONs.

Here is the folder of winning templates: [TEMPLATES FOLDER], sorted by concept (about 6 ads per concept).
My brand: [BRAND]. Logo and packaging locked, product photo attached.

First: accept EVERY template, whatever its concept, niche or product. Never skip one because it doesn't look like our angles.
Then sort the templates into concept families and work concept by concept.

For EACH template:
1. Describe its structure: layout, text hierarchy, visual format, type of proof, product placement.
2. Adapt it 1:1 to [BRAND]: same structure, same layout, our product in place of theirs.
3. Inside each concept family, spread the winning angles from the market research across the templates so the family covers every angle. One ad per template, with headlines from its angle's bank.
5. Extreme, aggressive, conversion-driven headlines, taken from the winning headline bank and then rewritten (never copied word for word).

Output: one JSON per creative using the ad-blueprint-gpt-image-2 skill.
- Image 1 = our product (to be reproduced identically).
- Image 2 = the template (layout reference only: not its text, not its product).
- Fields: concept, angle, headline_id, headline_final, gpt_image_prompt.
Volume = 1 ad per template. Start with 1 complete concept family and show it to me before running the rest.

Generate each JSON with GPT Image 2 (quality high, 1:1 square), Image 1 = our product, Image 2 = template.

Then launch a QC agent that checks EACH image at full resolution (never a thumbnail):
[ ] product identical to the reference (shape, colors, label, label text)
[ ] exact headline, no typos, no extra words, no duplicated text
[ ] offer, price and guarantee match our product sheet
[ ] structure faithful to the template (layout, hierarchy, product placement)
[ ] one-second test: problem, cause, product and guarantee are obvious to the avatar
[ ] no distorted hands or faces, no stray logos, no watermark
[ ] nothing banned by our product sheet

Any non-compliant image is regenerated using "change only X, keep Y". Give me a table: creative · status · fix applied.

Publish two pages on the VPS (<name>.machineroom.app, noindex, clean URLs):

1. Comparison: for each creative, template on the left and our ad on the right, grouped by concept, with the angle and headline under each pair.
2. Demo: a grid with ONLY our final images, no templates, no concept or angle names, no technical text. It's a demo link for someone outside the team.

Both pages must work on mobile. Give me both links.

Save the brand kit for [BRAND] in a single folder: [BRAND KIT FOLDER].
It must contain:
- the final market research report (angles, headline bank, video hooks, avatar);
- the product sheet (exact price, offer, guarantee, allowed claims, banned words);
- the product reference photo;
- the list of every creative already produced: concept · angle · headline · JSON · image.
This kit is the only source for every future batch.

New batch for [BRAND].

1. Reload the brand kit: [BRAND KIT FOLDER]. Do NOT research anything new, do NOT use TrendTrack: the angles, headlines, avatar and product sheet are locked.
2. Here are the new concept templates I picked: [NEW TEMPLATES FOLDER]. Accept every one of them, whatever the concept or niche.
3. Sort them into concept families, then work concept by concept.
4. Inside each concept family, spread the kit's winning angles across the templates so every family covers every angle.
5. For each template: copy its structure 1:1 (layout, text hierarchy, visual format, type of proof, product placement) and adapt it to [BRAND] on its assigned angle, with headlines from that angle's bank (rewritten, aggressive, conversion-driven). One ad per template.
6. Skip any combination already in the list of creatives made.
7. One JSON per creative with the ad-blueprint-gpt-image-2 skill, then generation with GPT Image 2 (quality high, 1:1) and QC by an agent (step 3).
8. Add the new creatives to the comparison site (each ad side by side with its template) and the demo link (step 4), and add them to the brand kit's list.

Start with the first concept family and show it to me before running the rest.