Creative Engine
AI ad-creative pipeline · for PC and mobile game studios

One concept in. Twenty on-brand ad variants out.

In about an hour. Built from your own art library, written in your own voice, rendered in every format your channels want. Creative Engine goes live on your stack, and your team runs it. For the feeds that eat creative: Meta, TikTok, AppLovin.

20 variantsper batch, from one spec sheet and one render
~1 hourhands-on time per batch, after setup
3 formatssquare, 4:5, and 9:16 from the same spec rows
100% yoursaccounts, templates, playbook. Nothing rented.

Creative is the bottleneck now

Paid UA runs on creative volume. The algorithm wants a constant feed of fresh variants to test, and most teams can produce a handful a week. Concepts sit in a design queue. Every scale-or-kill decision waits on the next batch. Meanwhile the board is asking what the AI plan is.

The obvious fixes don't hold. More briefs make the queue longer. Outsourced variant mills drift off-brand. And generic AI tools produce off-model characters and fake UI, the exact slop your players screenshot and post.

The bottleneck is throughput. Creative Engine is the throughput layer your team owns.

The output: a full batch

Tiny Tavern is a fictional idle game with a complete intake pack, run through the Engine end to end: brand rules, angle system, copy, templates, batch render, QA. These 20 came out of one spec sheet. Every stat and quote in the ads is fiction; the mechanics are the product.

Your batch is your art, your voice, your angles, built on your title during setup.

The console: where batches live and die

Every variant tracks through the loop: rendered, QA, in test, scaling, killed. Thresholds are yours, wired to whatever metric you run. This is a live demo with fictional numbers. Click around.

How it works

Step 1 · Setup · 2 weeks

The Engine learns your game

White-glove setup, included. The brand-context pack comes together from your art, brand rules, voice, and past winners. Layout templates get built on your brand, and the angle system tunes to what's running in your genre right now. Everything stands up on your accounts, and your team trains on real campaigns before setup ends.

Step 2 · Batch · ~1 hour

Concept to twenty variants

Pick a concept. The Engine generates angles and copy against the brand pack. Your team redlines, fills the spec sheet (one row = one ad, mostly dropdowns), and renders the batch in one command. QA checklist, then ship through your normal Ads Manager flow.

Step 3 · Loop · weekly

Results steer the next batch

Test results come back against your kill and scale thresholds. Winners get scaled and iterated, losers die fast, and next batch direction comes from the numbers, whatever metric you run: CPI, ROAS, payback.

What's inside

Brand-context pack

The single source every generation reads: asset library, brand and voice rules, past winners, genre hook patterns. The reason batch 40 still sounds like you.

Angle + copy engine

Angle matrix with rationale, then headlines, primary text, and CTAs per angle. Brand-voice guardrails with an eval pass, proven in daily production against a 26,000-product catalog.

Layout templates

3 to 4 components built on your brand: hook-top, character-led, screenshot-frame, social-proof. Each takes a headline, an art asset, a CTA, and a format size.

Batch render

One spec sheet drives everything. Twenty variants is one render command. Square, 4:5, and 9:16 from the same rows, named and versioned automatically.

Video lanes

Generated hooks, b-roll, and spokesperson clips through your own gen accounts. Your real gameplay capture where the game shows. Assembly cuts 15s and 30s versions plus motion statics.

QA gate

A human pass against a checklist: brand rules, thumb-stop legibility at feed size, platform-policy reject risk. Flags get caught before upload instead of after.

The rules it runs by

Your art, never AI characters

Characters, key art, and screenshots come from your library. Generated imagery stays bounded to backgrounds and texture plates. Where to push the line is your call; the Engine executes your creative direction.

Your dials, your numbers

Kill and scale thresholds wire to your metric and your numbers. Creative direction stays yours. Per-client settings all the way down, never hard-coded defaults.

Your accounts, your Drive

Everything lives on your tools: the sheet, the templates, the assets, the playbook. Walk away anytime and delete nothing; everything set up is yours.

What it promises: throughput

The Engine sells more shots on goal and faster scale-or-kill calls. Whether any single ad wins is your test loop's job, and no honest tool claims otherwise.

Questions studios ask

Will it draw our characters?
No. Your characters and UI come from your own asset library; AI writes, composes, versions, and assembles. Generated imagery is bounded to backgrounds and texture plates unless you decide otherwise.
What does our team actually touch?
Three things, two of which they already know: a spec sheet in your Drive (a spreadsheet: one row = one ad, about 30 seconds a row), a chat workspace that permanently knows your game, and one button to render. Uploading stays your normal Ads Manager flow. No engineer needed.
Does it touch the game itself?
Never. The Engine lives in your marketing stack. Data flows in through MCP connections from your attribution and ad platforms, so there are no custom integrations to maintain.
What about video?
Three lanes: generated hooks and spokesperson clips through your own gen accounts, your real gameplay capture wherever the game shows, and assembly that cuts hook + gameplay + endcard into 15s and 30s versions. Most winning game ads are hybrids of the three.
Is it self-serve?
Not yet. Every setup is white-glove: the Engine gets built onto your brand and accounts, and your team gets trained on it, in 2 weeks. It starts with a demo batch on your title.
What happens after setup?
Your team runs it; that's the design. An optional monthly retainer adds an optimization layer: weekly reads of your test results, scale-or-kill recommendations per variant family, and pipeline tuning as platforms shift.

Start with a demo batch on your title

Bring the game. The demo conversation starts from your public store page and what's already running in your genre, and turns into a batch plan for your title. Fixed-price setup, scoped on the call. Capped at 2 studios at a time.

30 minutes. If the bottleneck turns out to be wider than creative (reporting, campaign ops, research), that conversation exists too.