Skip to content

Visual Intelligence System · v1.0.0 · 2026-05-13

Drop images.
Get a per-platform strategy.

The substrate that connects your image catalog to your content calendar. Three asset tiers, nine platform personas, one workflow. Built on the systems already shipped at /watch, /workshops, /library, and /intelligence-system.

Open the runtimeSee the stack

The problem this solves

Generation without strategy is noise.

Every creator with NB2 access can generate a beautiful image. Few can answer the next question: which platform, which post, which hook, which variant. The gap between the image and the strategy is where most production capacity is lost.

VIS closes that gap with three commitments: every platform gets one persona; every asset routes through one of three tiers; every ship walks the same gate. The system refuses anything generic — the cost of refusing is small, the cost of shipping generic is the rest of your year.

The three-layer stack

What gets made. How it gets assembled. What stops bad work from shipping.

Each layer has a single job. The boundaries are deliberate — without them, the system collapses into one tool fighting another.

Asset Layer

What gets made. Two velocity tiers: NB2 for premium hero work; Higgsfield for the cinematic and product-grade volume tier. The authoritative engine registry — with live install status, cost, and the full menu (InfoGenius/NB2, Higgsfield, fal.ai, GPT Image 2) — lives at lib/gen/backends.ts (public at /studio/engines). The higgsfield-* entries below are roles, not installed skills; Higgsfield is reached via its MCP server.

  • nb-image

    Premium hero + book + album covers

    Hero imagery, book covers, album covers, anything that lives at the top of a page or stays on a feed for years.

  • higgsfield-product-photoshoot

    Brand-grade product and lifestyle photography

    Workshop landing visuals, course covers, papa-hub imagery, anything that needs studio-shot polish without a studio.

  • higgsfield-generate

    Cinematic stills and 5–15s video clips

    B-roll, cinematic cutaways, ad creatives, image-to-video animations, quote cards. The high-velocity volume tier.

  • higgsfield-soul-id

    Identity-faithful Frank-as-character training

    One-time training. After training, every higgsfield-generate call with --soul-id <id> produces faithful Frank-as-character output for B-roll where on-camera is impossible.

  • higgsfield-marketplace-cards

    Marketplace-compliant product listings

    Course Lemon Squeezy listings, Etsy/Amazon-style product cards, A+ content modules.

  • music-video-batch

    Lyric video generation across the 12k catalog

    Batch lyric videos for music releases. Composes with Higgsfield video for cinematic sections.

Composition Layer

How assets become videos. HyperFrames for caption-heavy short-form; Remotion for stateful long-form; the existing /talking-head-ship pipeline anchors brand standards.

  • hyperframes

    HTML-native composition with deterministic seek

    Caption-heavy shorts, Reels, TikTok, lower-third overlays, Lottie/After Effects imports.

  • remotion

    React-native composition with full state

    Long-form talking-head, branded YouTube uploads, anything reusing existing React components.

Gate Layer

What stops bad work from shipping. The vis curator decides which asset for which surface; visual-creation enforces the 6-step quality pipeline; brand-voice + design.md + taste.md are the final wall.

  • vis

    Curation — which asset for which page or platform

    Before generating: ask vis what already exists. Before publishing: ask vis whether the asset matches the page conversion goal.

  • visual-creation

    Quality pipeline — 6-step organic-first generation

    Every generation that will ship publicly. Council review and human approval are non-skippable.

  • brand-voice

    Voice + AI-tone refusal list

    Every caption, alt text, and on-image copy line.

The nine platform personas

One persona per platform. A profile that tries to be everything is read as nothing.

These are not aspirations — they are the operational rule used by content-social-distributor, the visual-intelligence skill, and the /visual-strategy command. Edit the substrate, every downstream surface updates.

LinkedIn

tech

AI Architect — the Oracle EMEA bridge story

First-person, technically precise, enterprise-fluent. The Oracle CoE framework translated to personal use, free.

Visual treatment
Cinematic stills with negative space for thread overlay, system diagrams in JetBrains Mono, carousel-format teardowns of real architectures.
Cadence
3 posts per week — 1 long-form thread, 1 short insight, 1 carousel teardown.
Asset source
nb-imagehiggsfield-generatehiggsfield-product-photoshoot
Forcing function
NLDigital 2026-05-19 / Madrid 2026-05-27 workshops drive demand thread frequency.

YouTube (long-form)

tech

The Builder — behind-the-scenes of shipped systems

Studio voice. Walk-throughs of ACOS, IIS, Watch OS, Workshop OS. Show the code, narrate the why.

Visual treatment
Talking-head A-roll + Soul-ID Frank for B-roll cutaways where you cannot be on camera + JetBrains code overlays.
Cadence
1 video every 10–14 days. Quality over rhythm. Each ships with a paired blog post.
Asset source
remotionhiggsfield-soul-idhiggsfield-generate

YouTube Shorts

tech

The Creator — one insight, sub-60s, captioned

Tight. One idea per short. The hook is the title.

Visual treatment
HyperFrames composition with caption-style block + Higgsfield 5–15s cinematic cutaway + brand lower-third.
Cadence
5 per week. Batch on Sunday from /watch/shorts pipeline.
Asset source
hyperframeshiggsfield-generate

TikTok

tech

The Creator — same soul as Shorts, different syntax

Faster, looser, lower-fi. Caption-on, sound-on, vertical native.

Visual treatment
HyperFrames composition tuned for TikTok caption ceiling + native font weight + Higgsfield Seedance for motion-heavy clips.
Cadence
5 per week, repurposed from Shorts cuts but with TikTok-native captions and sound choices.
Asset source
hyperframeshiggsfield-generate

Instagram

soul

The Aesthete — the 12k music catalog and generative art

Visual-first. Caption is liner notes, not pitch. Soul spectrum dominant — this is the music side of the brand.

Visual treatment
NB2 hero covers + Higgsfield carousel variants (4–6 frames per post) + occasional Reels from HyperFrames.
Cadence
4 posts + 2 stories per week. Album drops trigger Reel pipeline.
Asset source
nb-imagehiggsfield-product-photoshootmusic-video-batch

X

tech

The Thinker — tight insights, threaded teardowns

Condensed. Naval-density without the mysticism. Each tweet earns the next.

Visual treatment
Cinematic 4K stills as quote cards (Higgsfield Cinema Studio) + occasional Code-screenshot embeds + diagrams from /studio/visual.
Cadence
Daily. 1 thread per week.
Asset source
higgsfield-generatenb-image

Threads

tech

The Conversation-starter — lower-stakes, higher-frequency

Conversational. Questions over statements. Reply-bait done honestly.

Visual treatment
Mostly text. One quick Higgsfield generate per 3 posts maximum.
Cadence
5 per week. Repurposed X drafts that need more breathing room.
Asset source
higgsfield-generate

Bluesky

tech

The Live-thinker — mic-to-publish, the CIS MV1 surface

Voice-first. Captured via mic, lightly edited, published same-day.

Visual treatment
Quick-generate thumbnail per post. No deep production — speed is the point.
Cadence
Daily, per the CIS MV1 mic-to-publish workflow (Friday 2026-05-09 forcing function).
Asset source
higgsfield-generatenb-image
Forcing function
CIS MV1 first publish: 2026-05-09. Mic → transcript → Bluesky.

Spotify / Apple Music

soul

The Producer — the 12,000+ song catalog

No voice. The cover and the music carry it. Track titles are the only copy that matters.

Visual treatment
NB2 album/single covers (2K minimum, mimeType-derived ext per the 2026-04-25 regression rule) + Higgsfield cinematic visualizers for lead tracks.
Cadence
Per release. 1–4 covers per week depending on catalog batch.
Asset source
nb-imagehiggsfield-generatemusic-video-batch

The drop-images workflow

Five steps from raw image to ship-ready strategy.

Run via /visual-strategy, or dispatched automatically by the visual-intelligence-orchestrator agent on multi-image batches.

  1. 01

    Drop

    Paste images here, drop them in content/ingest/visual/, or upload a folder. The orchestrator reads each image multimodally.

  2. 02

    Analyze

    Subject, mood, lighting, composition, brand-fit, conversion potential, spectrum. Recorded as structured signals.

  3. 03

    Cross-reference

    IIS priorities, content calendar, hook-learn analytics, existing /vis registry. The image lands inside your strategy, not next to it.

  4. 04

    Recommend

    Per image, per platform: which persona, which hook pattern, which variants needed, which tool to fill the gap.

  5. 05

    Ship or fill

    If the image is ready: route through visual-creation gate to publish. If gaps exist: NB2 or Higgsfield queue with brand-tuned prompts.

The runtime

One command. One skill. One agent.

VIS is operated through the existing FrankX surfaces — Claude Code today, any MCP-aware client tomorrow. Nothing custom; everything composable.

Slash command

/visual-strategy

Run on a batch of images. Returns per-image, per-platform recommendations and a gap-fill plan.

Skill

visual-intelligence

Composes vis + brand-voice + social-media-strategy + nb-image + Higgsfield. Auto-triggers on visual-platform requests.

Agent

visual-intelligence-orchestrator

Multi-image batch handler. Sub-dispatches to vis, brand-voice, nb-image, and Higgsfield in parallel.

Example invocation

# In Claude Code, point at a folder of images
/visual-strategy content/ingest/visual/2026-w19/

# Or paste images directly into the conversation and ask
"Run visual-strategy on these. Bias toward LinkedIn (NLDigital workshop)
and Spotify (Friday album drop). Use Higgsfield for variants needed."

# Output: per-image analysis + per-platform plan + gap-fill brief
# Drafts land in content/staging/visual/<batch-id>/

How it composes

VIS is not a new system — it is the connective tissue.

The asset, composition, and gate layers are already shipped. VIS adds the typed substrate, the persona matrix, and the workflow that ties them together.

Watch OS

Receives shorts cuts; feeds B-roll back into VIS for Reels variants.

Agentic Content Officer

The talking-head pipeline; VIS supplies cinematic cutaways via Higgsfield Soul ID.

Workshop OS

Receives workshop hero imagery; per-attendee brief visuals via product-photoshoot.

Library OS

Cover generation routes through nb-image; VIS curates which book gets a Higgsfield carousel.

Investment Intelligence

VIS analyzes Strategy Session diagrams for blog-post visualization.

FrankX OS

VIS appears as the 9th OS module; cross-linked from every shipped surface.

Ready to drop images?

Open Claude Code, run /visual-strategy, and pass a batch of images. Or paste them straight into the conversation.

See VIS in the FrankX OS

v1.0.0 · shipped 2026-05-13 · MIT-aligned with the Library OS pattern