Higgsfield Content Factory

Higgsfield Content Factory — Campaign Orchestration Pipeline

A 5-stage pipeline that runs a full ad campaign on top of Higgsfield Marketing Studio: Research → Plan → Generate → Publish → Report. Where higgsfield-marketing-studio is the engine reference (what the presets, parameters, hooks, and avatars are), Content Factory is the orchestration layer that drives a whole campaign turn by turn — researching trends, planning a dated content calendar, generating in batches, scheduling to ads, and reporting cost savings.

Translated from Adil Aliyev's higgsfield-content-factory source skill, per the v3.7.13 / v3.7.16 translation precedent. All Marketing Studio API ground-truth — preset slugs, parameter schema, hook/setting/avatar handling — lives in ../higgsfield-marketing-studio/SKILL.md and is NOT restated here. This sub-skill carries the pipeline scaffolding the marketing-studio sub-skill deliberately deferred. The publish + cost-report tail (Stages 4–5) lives in the companion publish-and-report-workflow.md satellite.

---

1. Two operating principles

Content is UGC-first. Every campaign defaults to a heavy share of UGC-family content — talking-head clips, honest reviews, ASMR close-ups, unboxing reveals. Hyper Motion, TV Spot, and Wild Card stay available but off the default mix. Every video idea must be producible inside one live Marketing Studio preset within its duration cap (see marketing-studio § 3 / § 7 for the canonical preset list and limits).

The interaction is button-driven. Every clarifying question is asked as a 2–4 option button choice (one round-trip), never a free-form "type your answer" for navigation, confirmation, or routing. Free-form typing is reserved for content the user must originate — and even then, always offer a smart default they can accept with one click. The only non-button steps are attaching a product image or pasting a product URL (both click-based file actions).

User-facing language rule (HARD). The user is not a developer. Do not narrate technical mechanics — no MCP tool names, no UUID resolution, no "running parallel searches," no enhanced-prompt internals. All of that runs silently. Send one plain-language stage banner when each stage starts, a brief "Stage N done — [deliverable]" between stages, then the next banner. The user sees a clean narrative arc, not a tool log. If the user asks "how does it work under the hood," then explain.

Stage banners

Send one banner per stage, in plain language (no tool names):

---

2. Onboarding — one message, all buttons, no pauses

Ask the onboarding questions in a single button round-trip at the start. Do not ask them sequentially; do not pause between onboarding and the product request.

• A — Higgsfield connection: "Yes — connected" · "Not yet — I'll connect now" · "Skip — research only."
• B — Starting stage: Stage 1 full pipeline (needs a product) · Stage 2 build content plan (have a brief) · Stage 3 generate now (have a plan) · Stage 4 schedule (content ready). A product image with no other context defaults to Stage 1.
• C — Video volume: 50 · 100 (recommended) · 150 · 200 · Other (any number). Store as [VIDEO_COUNT].
• D — The product: "Attach your product image OR drop a URL — that's all I need to start." Skip D if a product is already attached. For Stage 3/4, swap D for "drop your existing content plan."

Send A + B + C as buttons and the product-attach prompt in the same message; once answered, proceed straight to the chosen stage with no extra confirmation.

Compute the per-format split silently the instant [VIDEO_COUNT] locks in — do not announce the breakdown here. It surfaces later, woven into the Stage 1 brief as a consequence of what the research showed, so it reads as a trend-driven choice rather than a config card.

---

3. The campaign mix — 5 UGC formats (even split)

Every campaign distributes evenly across five viral UGC formats. Cinematic secondary types (Hyper Motion / TV Spot / Wild Card) stay off-default — used only when the user explicitly asks.

Allocation math. per_format = floor(VIDEO_COUNT / 5). If VIDEO_COUNT doesn't divide evenly, distribute the remainder one-per-format from format 1. Examples: 100 → 20/20/20/20/20; 50 → 10×5; 17 → 4/4/3/3/3.

This is the default mix and is user-overridable. The default share is a craft choice (UGC-first), not a platform rule.

Format → preset routing

Each format maps to a Marketing Studio preset. Route preset selection through the canonical mechanism documented in ../higgsfield-marketing-studio/SKILL.md § 3 (slug naming is canonical there; routing mode lives on a separate MCP call). At a glance:

• Formats 1 (Entertainment), 2 (Street Interview), 5 (ASMR) → the UGC preset; what differentiates them is the system hook, the setting, the audio flag, and the prompt content — not a separate preset.
• Format 3 (Unboxing) → the Unboxing preset.
• Format 4 (Product Review) → the Product Review preset.

Do not hardcode hook/setting/avatar UUIDs or invent a generate_video.mode slug. Enumerate hooks/settings live (marketing-studio § 4 live-enumeration discipline) and pass the resolved IDs at generation time.

Concept seeds (vary within each format)

No two videos in the same format should be the same concept. Pull from seeds like: blind taste try, "$100 to try it" street challenge, will-it-pour (Entertainment); "rate it out of 10," "sing for the product," blind opinion → brand reveal (Street Interview); trio reveal, ribbon-pull solo drop, subscription-box drop (Unboxing); two-ingredient test, fridge-ranking, 7-day diary (Product Review); macro cap-unscrew + glug pour, condensation slide, spoon-clink ice-drop (ASMR).

ASMR is a style, not a preset

Render ASMR as the UGC preset with audio on, an intimate low-noise setting (Kitchen / Bathroom / Bedroom), usually no system hook, and a prompt focused on close-up handling sounds (cap unscrew, pour, glass clink, condensation). No talking; caption-only or silent open.

---