15 Sept 25 
AI can produce a hundred ad variants before lunch. It can also make your brand look like everyone else's. Here's the production checklist we use to get the volume without the plastic sheen.
A small confession to open with.
A while back, we reviewed a batch of AI-generated hero images for our own blog. Individually, each one looked lovely: soft pastel sky, fluffy clouds, cute floating icons. Then we put six of them side by side and had an uncomfortable realisation. They all looked like the same image wearing different hats. Same clouds. Same sky. Same cheerful little icons that belonged in a children’s app, not an agency portfolio.
We’d fallen into the exact trap we warn clients about. The machine gave us what we asked for, at speed, at volume, and it was quietly making us generic.
That experience is why this checklist exists. We’ve spent years building creative automation pipelines, including one that produced adapted campaign assets across 45,000+ stores for Amazon Easy, and the single biggest lesson is this: AI solves the production problem, not the taste problem. Taste still has to come from somewhere. Here’s how we make sure it does.
The Problem
Why AI Creative Drifts Toward “Generic”
Every image model is trained on the averaged output of the internet. Ask it for “a professional ad for a skincare brand” and it gives you the statistical middle of every skincare ad it has ever seen: glowy skin, white bathroom, eucalyptus sprig. Technically flawless. Instantly forgettable.
The tells are consistent enough that consumers now spot them on instinct:
- Waxy, poreless skin and teeth that glow slightly
- Text that dissolves into alphabet soup at the edges
- Compositions so symmetrical they feel embalmed
- That over-saturated “digital painting” light no camera produces
- And the subtlest one: sameness across your own campaign, where every variant has the same soul because they came from the same prompt
None of these are reasons to avoid AI creative. They’re reasons to run it through a process. Scroll through the Meta Ad Library for any category you compete in and you’ll see both futures: brands drowning in interchangeable AI slop, and brands quietly using the same tools to out-produce everyone while still looking unmistakably like themselves.
The Solution
Our 7-Point Production Checklist
This is the actual pass every AI-assisted creative goes through before a client sees it, let alone an ad account.
✅ 1. Brief the machine like you’d brief a designer
“Make an ad for our sale” gets you slop. Our prompts specify audience, emotional temperature, composition, lighting, palette (as hex codes), and, critically, what to avoid. A prompt without exclusions is a prompt that drifts. Ours ban the clichés of whatever category we’re working in, by name.
✅ 2. Anchor to real brand assets
Every generation references actual brand material: product shots, existing photography, the brand’s precise colour values. AI extrapolating from your brand looks like you on a productive day. AI extrapolating from nothing looks like everyone.
✅ 3. Keep an art director in the loop, always
At Identiti this is non-negotiable. A designer reviews every batch, kills 70 to 80 percent of it, and refines the survivors. The machine proposes; a human with taste disposes. The moment nobody with design judgement is looking, quality decays within weeks. We’ve watched it happen.
✅ 4. Never ship the first render
The first output is a draft, the way a first sketch is a draft. Our floor is three iteration rounds: generate wide, select hard, regenerate narrow. The gap between render one and render four is usually the gap between “obviously AI” and “wait, how did you produce this much creative?”
✅ 5. Do type and logos outside the model
Image models still mangle words. Even when they don’t, they can’t respect your brand’s typography. All copy, logos, CTAs, and legal lines get composited in design tools afterward, set in the brand’s real typefaces. This one habit removes the single most common AI tell.
✅ 6. Run the squint test, as a set
Line up the whole batch and squint. If five variants blur into one shape, you have one ad and four clones, and your creative testing will tell you nothing. Variants must differ in concept (different hook, different composition, different emotional angle), not just in background colour.
✅ 7. QA the details before export
Hands and fingers. Reflections. Background faces. Jewellery that merges into skin. Product labels that almost-but-don’t say your product’s name. Thirty seconds of zoomed-in inspection per asset, every time. The most expensive AI mistakes are the ones a stranger screenshots.
What This Unlocks
Here’s the honest economics. A traditional shoot-and-design cycle produces maybe 5 to 10 ad concepts a month. An AI pipeline with this checklist bolted on produces 50 to 100 tested-worthy variants in the same window. That means your media buyer finally has enough creative to let the algorithm find winners instead of fatiguing one “hero ad” into the ground.
Speed without taste makes noise. Taste without speed loses auctions. The checklist is how you keep both.
AI solves the production problem, not the taste problem. Budget for taste.
If you’d like the checklist applied to your own ad account, or you’re curious what a creative automation pipeline would look like for your brand, that’s precisely the kind of engagement we love.