The Static-Ad Template System: Angle × Template × Product

The Static-Ad Template System

TL;DR: A finished static ad is the product of three independent axes: ANGLE (what to say — from audience research), TEMPLATE (how to lay it out — a brand-agnostic layout skeleton), and PRODUCT (whose brand — palette, fonts, product imagery). Separate them and ad production becomes a fill operation instead of a from-scratch creative act. Two complementary moves run the system: a foundry deconstructs existing ads into reusable, catalogued templates (compressing many ads into few parametrized layouts), and an apply step pours an angle + a brand into a chosen template to render a finished ad in every aspect ratio. The durable asset is the template library, not any single ad — and the discipline that makes it work is compress, don’t fork: six similar ads become one parametrized template, not six entries.

Why separate the three axes

Most ad production fuses three different decisions into one act of “making an ad,” which is why it doesn’t scale and why AI generation of it feels generic. The system pulls them apart:

Axis	Question it answers	Where it comes from
Angle	What do we say? (the hook, promise, proof)	Audience research — the unit table from marketing/evidence-graded-audience-research / a `units.json`
Template	How is it laid out? (composition, slots, focal order)	The brand-agnostic template catalog
Product	Whose brand? (colors, fonts, product photo)	A per-brand kit (palette, type, voice, assets)

The same template renders any brand; the same angle can be cast into several templates; the same brand flows through every template. Because the axes are independent, one improvement compounds across the whole grid — a better template lifts every angle×brand that uses it; a sharper angle lifts every template it’s cast into.

This is the staged-compiler durable-vs-regenerable split applied to static creative: the template library is the durable, compounding layer; the individual rendered ad is the cheap, disposable output.

The two moves: foundry → apply

Foundry — build the template library (deconstruct → catalogue)

The foundry takes an existing static ad (yours or a competitor’s) and abstracts it into a brand-agnostic template: a layout skeleton, named swappable slots, generation rules for the base image, a “when to pick this” note, and a status. It reuses the deconstruction half of seo/ai-template-casting-workflow and the structural-not-surface discipline — the template captures the formula (composition, focal hierarchy, slot structure), never the skin (glossary/creative-formula-vs-creative-skin).

The load-bearing discipline is dedupe — compress, don’t fork. Ten source ads do not become ten templates. If six ads share one layout with different photos and copy, that’s one parametrized template with six fillings, not six entries. (The first real run compressed 10 source ads into 4 templates.) A library that forks per-ad isn’t a library — it’s a screenshot folder.

Apply — render a finished ad (template × angle × product)

The apply step is the inverse: pick a template that fits the angle, pull the actual copy from the angle research (it doesn’t invent the message), pour in the brand kit, generate or drop in the base photo per the template’s gen-rules, fill the slots, and render. Output is the finished ad in all aspect ratios (4:5 / 1:1 / 9:16), with per-ratio position overrides so alignment survives the reflow.

Pulling copy from the research rather than re-generating it is what keeps the output grounded — the apply step is an assembler, not a copywriter. Garbage angle in, polished-but-empty ad out; the template can only lay out a message, not supply one.

Anatomy of a template

Each catalog entry is one layout, fully brand-neutral:

Layout skeleton — the composition (e.g. “cream copy-band in the top safe zone over a full-bleed lifestyle photo”).
Slots — the named fill points (headline, cta, quote, attribution, photo, chips[]…), some optional.
Focal order — the intended eye-path through the slots (the focal hierarchy).
When to pick — which angle type this layout serves (aspiration/identity, segment testimonial, interactive pattern-interrupt…).
Gen-rules — how to produce the base photo so it leaves room for the overlay (e.g. “keep the top third calm and low-detail for the copy band; real not stocky; no faces needed”).
Status lifecycle — experimental (untested) → live (proven in a results read) → retired (kept for reference, out of rotation).

For layout-style templates (typographic overlays on a photo), the system also emits a working compositor: a brand-agnostic drawing runtime + a per-brand kit + a content JSON, so idea → JSON → base photo → render is deterministic and repeatable across aspect ratios — not a fresh prompt every time.

Where it sits among the wiki’s creative methods

This is easy to confuse with the reverse-engineering cluster. The distinction is persistence:

Method	Input → output	Persists?
ad-alchemy / seo/ai-template-casting-workflow	one competitor ad → 5 one-shot variation prompts for one product	No — ephemeral, per-run
Static-ad template system (this page)	a catalogued template × an angle × a brand → a rendered ad	Yes — the template library is a durable, reusable asset
tools/scenario-compiler	a research unit → a prescriptive production package	The package; templates feed its visual rows

Casting is the one-shot move; the template system is the standing library that casting outputs can graduate into. They share the same deconstruction front-end and the same formula-vs-skin discipline — this page reuses them, it doesn’t restate them.

It’s also a sibling in shape to marketing/slideshow-pattern-design (a catalog of reusable patterns mapped to when-to-use) — same “named, catalogued, pick-by-situation” structure, different surface (paid static vs organic slideshow).

Honest limits

Templates lay out a message; they don’t create one. The system’s quality ceiling is the angle research feeding it. A library can’t rescue a weak angle — it can only make a weak angle look finished, which is worse.
A template library standardizes — and standardization can flatten. marketing/andromeda-era-creative-strategy is explicit that diversity must live at the formula level: 10–15 genuinely different concepts, because the delivery engine suppresses near-duplicates. A library of one over-used template produces sameness the auction punishes. The library is the floor of consistency, not a substitute for concept diversity.
The specific templates are unproven. Every template in the current catalog is status: experimental. The methodology is durable; the catalog is not yet validated — a template earns live only from a real CTR/CPC read, and only live templates should ever be treated as recommendations.
The compositor covers layout-style statics (typographic overlays). Fully-photographic concepts still depend on the gen-rules + an image model, with the usual fidelity caveats (marketing/ai-product-image-generation); fully-AI creative that looks AI-made carries a measured trust penalty (glossary/human-anchored-ai-multiplication).

Key Takeaways

A static ad = angle × template × product, three independent axes. Separating them turns production into a fill operation and makes every improvement compound across the grid.
Two moves: a foundry deconstructs ads into brand-agnostic templates; an apply step pours an angle + brand into a template and renders all aspect ratios.
Compress, don’t fork — many similar ads collapse into one parametrized template; a per-ad library is just a screenshot folder.
The template library is the durable, compounding asset; the rendered ad is disposable output (the staged-compiler split applied to static creative).
The system lays out a message; it can’t supply one — quality is capped by the angle research. And specific templates stay experimental until a real results read promotes them.

seo/ai-template-casting-workflow — the deconstruction → variation half this reuses (one-shot casting; the template system persists its output)
seo/surface-vs-structural-mimicry — why a template captures structure, never skin
glossary/creative-formula-vs-creative-skin — the formula-vs-skin distinction the foundry runs on
marketing/slideshow-pattern-design — the sibling pattern-catalog (organic slideshows); same catalogued-pick-by-situation shape
marketing/andromeda-era-creative-strategy — why the durable unit is a reusable template library, and why formula-level diversity is mandatory
automation/staged-compiler-pattern — the durable-vs-regenerable architecture this instantiates
tools/scenario-compiler — the production-package compiler the templates feed
tools/ad-template-foundry — the skill that builds templates (deconstruct → catalogue)
tools/ad-template-apply — the skill that renders ads (template × angle × product)
marketing/evidence-graded-audience-research — where the angle axis comes from
glossary/focal-hierarchy — the eye-path each template’s slot order encodes

Sources

Primores internal skills (18-ad-template-foundry, 19-ad-template-apply) and the nutriguide-ad-pipeline/ad-templates/ library — brand-agnostic catalog (10 templates across photographic-DR and UI lanes), toolkit.py drawing runtime, per-brand brandkit.py, PIL compositors, compose.py. Built and dogfooded June 2026; templates currently status: experimental.

By Andrej Ruckij