The system is the work. I build creative production pipelines for AI video, product UI, and real-time 3D.

Standard AI generation breaks character and environment consistency across sessions — built a reference architecture that pins visual rules so any generation step produces on-system results. Reduced shot production cycle from brief to ready-for-edit across 3+ campaigns. Reusable reference pack held visual consistency across 20+ characters and 4 environment systems.

Reduced repetitive manual production work by systemizing the full path from 3D render to final video. What previously required a full day of manual work per asset now runs as a repeatable, data-driven pipeline capable of generating large volumes of consistent animated content while preserving creative control over characters, visuals, motion, audio direction, and final composition.

Turned a brand concept into a scalable design system covering logo direction, visual rules, style guide documentation, reusable templates, and consistent application across all documents and visual touchpoints. The system-first approach — rules before assets — means any team member can extend the brand without breaking it. Rejected the common logo + colour palette approach in favour of a documented token layer that governs every output.

Style drift is the usual problem with AI-illustrated books: page 12 stops looking like page 1. Here the look lives in the graph, not the prompt. The styler template, checkpoint and LoRA weights are fixed, so only the subject changes between pages. I also revived and cleaned up older workflows: fixed missing nodes and models after node-pack updates, removed dead branches, and laid each graph out in four readable stages.

Every take sits next to the prompt that made it, and each decision is recorded in the sheet production already works from. There is no backend and no build step: it is one HTML file the team can open anywhere. The screenshots use invented demo data; the real production is under NDA.

GA4, Search Console and LinkedIn share one date range with period-over-period comparison, and each post is tied to its campaign and tracked link, so planning and results sit side by side. The screenshots use a synthetic dataset; the live instance is private.