node-based -> agentic creative pipeline
Building for Meta AI's production systems: how do you enable artists to ideate and productionize visual content for every seasonal moment, globally, at speed? This configurable node-based creative pipeline (e.g. ComfyUI, Figma Weave) systematizes cultural research, concepting, ranking, and prompt engineering. Before: a single artist spent 1-2 weeks iterating on one concept. After: one artist configures a concept in 1 day => producing +33% CTR. The node workflow solved speed, then evolved into an agentic system that further harnessed and QA'ed each technical artist's process.
Deeply Engineered Prompts
50K+ character system prompts per node encode domain expertise no generic LLM call reproduces. Purpose-built for fun and culturally relevant content.
Multi-Model Intelligence
Routes between Claude Sonnet, Gemini Pro, and Llama Maverick for reasoning; Midjourney, Flux Kontext, and Faceswap for visual output.
Scaled with Speed
From 1-2 weeks iterating on one concept to configuring a concept in 1 day. Same artist, configurable pipeline, global seasonal coverage.
production visuals
agentic execution
But rigid graphs still limited artists: black-box nodes, no inter-run memory, every artist solving output issues differently. When agentic systems matured internally, the pipeline evolved again. The new agentic pipeline uses a unified execution document that blends natural language, code, system prompts, and artist configs in a shared drive to run entire multi-node workflows in one command, with agent-guided iteration that gives artists a true collaborative creative partner. Checkpoint resume to save LLM bandwidth, art direction assistance, full provenance logging, and cross-concept comparisons. The system improves with use.
Chat to Create
Artists chat with an ai agent that understands how to run the entire node pipeline and to fix issues.
No-Code Mode
Artists at any technical level can systematize creative visual prompting.
Built for Prototyping
Artists working out of a shared cloud drive can iterate on each other's work.
seasonal moments