Step 12 · Product layer · Product layer · Marketing Factory ENPT
Alembic Complete Visual Course

Marketing Factory

Turn verified business signals into go-to-market assets and versioned manifests.

Read the plain version, or open the technical layer on any section.
1

Signal → assets


@alembic/marketing-factory consumes a BusinessSignal, loads marketing skill contexts (product-marketing, copywriting, launch, pricing, CRO), builds positioning and copy, generates creatives via the Higgsfield CLI, scores them with a virality predictor, and writes a versioned AssetsManifest.

Offline mode uses deterministic templates and a fake Higgsfield seam. Online mode shells out to the real higgsfield CLI.

Think of it like… an ad agency in a box: brief in, positioning, copy, visuals, and a campaign report out.

Under the hood

The pipeline is runMarketingFactory: validate signal → load skills → build context → build copy → create Higgsfield generation → wait → score outputs → assemble manifest. The manifest is content-addressed by hashing signal + context + copy + creatives. runMarketingBatch runs sequentially over many signals.

2

In one picture


Signal Skills Copy Creatives Manifest
Business signal flows through skills, copy, creatives, and into a versioned manifest.
3

In the code


# offline run from a signal JSON
alembic marketing signal.json --offline

# online run bound to a brand soul
alembic marketing signal.json --online --soul-id my-brand

Implementation notes

The CLI reads the signal file, ensures marketing skills exist in <dataDir>/skills/marketing, builds a Higgsfield client (fake or real), and calls runMarketingFactory. The manifest is written to <dataDir>/marketing/<signalId>/manifest.json.

4

Try it


Create a signal.json with a BusinessSignal and run alembic marketing signal.json --offline. Inspect the manifest.

Provenance: Why does the manifest keep rejected creatives?
5

Quick check


What does the marketing factory produce as its final artifact?