Watch the walkthrough video: https://youtu.be/KCMrd_DM5uY

Marque is an agentic brand studio: one continuous pipeline from brand intake to published artifacts — profile → strategy board → moodboard → campaigns → shoots → sites → brand book.

Ask an AI what festival-going metalheads like and it will answer confidently — from vibes in its training data. That is the problem Marque is built to remove: it does not ask the model what an audience wants, it asks Qloo.

What it does

  • Brand DNA intake: scrape a URL or brief-chat a profile into voice, palette, and audience — enriched with Qloo taste-graph affinities.
  • Strategy boards: campaign plays generated against the audience, backed by logged Qloo calls; every board shows strategies, channels, and estimated reach with measured figures — none intuited.
  • Audience reads run as live calls to the Qloo taste graph before a word is written. Scores (e.g. 86 on audience fit) are computed from affinities, not generated to sound plausible.
  • Creatives: each rendered creative is scored against the graph and stamped with the model that produced it. The model makes the pixels; it does not make the claims.
  • Evidence drawer: every score produces its receipt — entities, tags, latency, and the affinities the audience verifiably over-indexes on. Nothing is hand-waved.
  • Downstream outputs: moodboards, photoshoots, generated sites, and a brand book all inherit the same measured context.

How it was built Express API + React/Vite SPA over a single JSON state document; an in-memory job runner drives long work (moodboards, shoots, video). Qloo's API supplies the taste graph — entity resolution, tag affinities, and demographic weighting — wired into profile building, strategy generation, and creative scoring. Local SDXL renders the imagery; FFmpeg + Playwright film the walkthroughs.

The point Generation where it belongs, measurement where it matters. One pipeline — receipts attached.

Built With

Share this project:

Updates

Submission history