Palasury — The Creative Engine

(01)The problem

Most beauty brands don't have an ad problem. They have a strategy problem.

Producing ads has never been cheaper. Deciding which ads are worth producing still takes a senior strategist most small brands can't hire.

One winner, carrying everything

The same creative runs for months. Frequency climbs, costs climb, and the account is really testing one angle.

ii.

Tests without a hypothesis

New ads get made, but nobody wrote down who they're for, which objection they answer, or what the test should teach.

iii.

The strictest category on Meta

Beauty ads get rejected for before-and-afters, personal-attribute language and medical-sounding claims. Most of that risk is decided in the brief.

(02)Process

From connected data to a brief you can see.

The founder spends about ten minutes answering questions. The agents handle the rest, and every output stays editable before anyone spends a dollar on production.

STEP 01↳

Connect

Add the website and product URLs, connect Meta ads with read-only access, and drop in reviews or past briefs.

IN · site · products · ad account

STEP 02⌕

Agents research

Parallel agents read the brand, its reviews, its live ads and the category in Ad Library, then compare notes.

OUT · teardown · gaps · customer language

STEP 03?

Founder intake

The intake agent asks only what the data can't answer: margins, legal limits, the things the founder won't say.

OUT · context the data doesn't hold

STEP 04≡

Strategic brief

Angles, personas, hooks, formats and a test plan, checked against Meta's beauty ad policies before delivery.

OUT · brief · test plan · compliance

STEP 05◐

Concept prompts

Every concept ships with image prompts tuned for Nano Banana, GPT Image, Midjourney and Flux. Generate a rough in minutes, then brief designers and videographers with a picture, not a paragraph.

OUT · 4 models · per concept

Step 05 · in detail

See the concept before you pay to shoot it.

A brief describes an idea. A rough image shows it. Each concept in the brief comes with a prompt the team can paste into the image model they already use, so the founder, the designer and the videographer are looking at the same thing from day one.

01One prompt per concept, adapted to each model's syntax and strengths
02Product, lighting, composition and format locked to the brief
03No text rendered by AI. Typography is set by a designer, every time
04Roughs are for alignment and testing logic, not final ads

Concept S2 · Routine subtraction

"Four steps. Retired."

Tests whether "less work, same results" earns more attention than a results promise.

Concept prompt4:5 · no text

Overhead flat-lay of a travertine stone tray with five carved slots. Four slots empty, the fifth holds an amber glass dropper bottle with a gold collar and a blank cream label. Matte charcoal surface, soft window light from upper left, warm amber highlights, editorial product photography, shallow shadows, 4:5.

Nano BananaGPT ImageMidjourneyFlux

(03)The agents

Six specialists. One brief.

Each agent owns one job and hands structured output to the next. It's the division of labor a good creative team already uses, compressed into one run.

Agent 01Research

Research agent

Reads the brand like a new strategist on day one.

Website, product pages, ingredients
Reviews and recurring customer language
Competitors in Meta Ad Library

Agent 02Performance

Performance agent

Reads the ad account like a media buyer.

Fatigue: frequency, spend concentration
Hooks and formats that hold attention
Angles that were never tested

Agent 03Intake

Intake agent

Turns gaps in the data into a short founder interview.

Asks only what the data can't answer
Captures legal limits and brand no-gos
Confirms the priority buyer

Agent 04Strategy

Strategy agent

Writes a brief a team can execute without a follow-up call.

Three angles, each tied to a buyer and an objection
Hooks, formats and creator direction
One-variable test plan

Agent 05Compliance

Compliance check

Reviews every line against Meta's rules for beauty ads.

No misleading before-and-afters
No personal-attribute language
No medical or guaranteed claims

Agent 06Concepts

Concept prompt agent

Turns each concept into prompts the team can run today.

Nano Banana, GPT Image, Midjourney, Flux
Product and lighting locked to the brief
Rough first, production second

Agents advise. People decide.

Nothing ships automatically. Every brief is a draft until the brand edits and approves it, and the engine never touches a live campaign.

(04)Sample output

The workflow already exists. It runs by hand today.

Before writing the software, the full process runs manually: a fatigue diagnosis, three angles, a test batch of four statics, and a creator brief. This study is one result.

Static ad: Your retinol didn't fail. Its delivery did.

S1 · Angle 01 · 4:5

Does an education hook beat the outcome promise?

S2 · Angle 02 · 1:1

Does "less work, same results" earn interest?

S3 · Angles 01–02 · 4:5

Does a claim-free texture stop the scroll?

Static ad: Strong skin doesn't have an age.

S4 · Angle 03 · 4:5

Does identity language beat category language?

Strategy study page 1: diagnosis and teardown

Strategy study page 4: rationale and projected test plan

Strategy study page 5: UGC creator brief

Disclosure

Self-initiated concept study. VÉLARE is a fictional brand created for this study, no client relationship implied. All performance discussion is projected test logic, not measured results. Product imagery was AI-generated and checked by hand; all typography was set by hand.

(05)Founder

Built by a media buyer, not a prompt collector.

I'm Fatih Palas, founder of Palasury. Before this, I ran paid growth for a premium DTC brand, where Meta drove most of the online revenue and every creative decision had a budget attached to it.

The lesson was expensive and simple: winning ads aren't the prettiest ones. They're built on a clear buyer, a clear objection and a clear reason to test. That thinking is what small beauty brands can't afford to hire full-time.

The Creative Engine packages the process I already run by hand: research, account teardown, angle strategy, brief, compliance check and concept prompts. The product makes it available to brands that will never have a strategist on staff.

170+

ad creatives tested on Meta with spend-based kill-and-scale rules

4.36

ROAS on the top creative

~80%

of the store's online revenue came from Meta ads

~$140

average order value, premium DTC

Figures from my previous role leading paid growth at a premium DTC eyewear brand, 2025–2026.

(06)Built on Claude

Agents that reason, not templates that fill in blanks.

Strategy work is long-context reading, judgment and careful writing. That's why the engine is being built on Claude.

How it's built

Claude API and the Claude Agent SDK for the multi-agent pipeline, with tool use for web reading and data access
Long context to read a brand's full site, reviews and ad history in one pass
Meta Marketing API with read-only permissions (planned, subject to Meta app review)
Structured hand-offs, so every claim in the brief traces back to a source

Principles

Read-only by design. The engine never edits campaigns or spends money
Human approval on every brief
Compliance built into the brief, not patched on after a rejection
Honest about AI. No fake results, no ROAS promises

(07)Roadmap

Where it stands today.

Validate the workflow by hand first, then automate what's proven.

Now

Workflow validated by hand

The full process runs manually with Claude and is launching as a service for beauty brands.

Agent pipeline prototype

Research, intake, strategy and concept agents on the Claude API, tested against real brand data.

Then

Private beta

A small group of US skincare brands using the panel, every brief reviewed by hand.

Later

Meta integration

Read-only ad account sync and a loop that learns which angles actually won.

(08)FAQ

Straight answers.

Is The Creative Engine live?

Not yet. It's in development, and the same workflow runs by hand today. Brands that request early access will be first in the private beta.

Who is it for?

US skincare and beauty DTC brands that already run Meta ads but don't have an in-house creative strategist. Not medical, not MLM, not enterprise.

What are the concept prompts for?

Fast alignment. Teams paste them into Nano Banana, GPT Image, Midjourney or Flux to see a rough of each concept, then use it to brief designers and videographers. They aren't finished ads, and AI never renders the text.

Does it run my ads or manage my account?

No. It reads data and writes strategy. It never edits campaigns, changes budgets or spends money.

Do you guarantee ROAS?

No. Nobody honest can. The engine gives you well-reasoned angles and a clear test plan. The market decides what wins.

How is AI used?

Openly. Claude powers the agents, and AI image models are used where they make sense. Strategy decisions and approval stay with people.

Creative strategy
for beauty brands, run by agents.

The Fatigued Bestseller