Cole Medin

I Turned Claude Code Into a Complete Video Generation System (with Archon)

⏱ 14 min video · 3 min read12 Jul 2026Worth watching

TL;DR

Cole Medin demos an AI-powered video generation pipeline that combines his open-source orchestration tool Archon with the Higgsfield video platform to automatically create product marketing videos (renders and UGC-style ads) from a product catalog. The system uses parallel agent workflows to scale video generation across dozens of products without overwhelming a single coding agent.

Key points

Archon (23,000+ GitHub stars) is used here as a general agentic workflow orchestrator, not just for AI coding — demonstrating its flexibility for content creation pipelines

Higgsfield is the video generation platform used to produce both product render videos and UGC-style ad videos via a CLI, accepting branded reference images for accuracy

Two Archon workflows drive the system: an 'explore' workflow that generates and validates image concepts first (to save credits), and a 'render' workflow that produces final videos from approved concepts

The explore workflow fans out 5 parallel worker agents to drain a product queue; the render workflow uses 3 parallel workers — each claiming one product at a time from a shared queue using a RLAIF loop pattern

A human-in-the-loop approval step sits between the two workflows: an AI scores and filters images automatically, then a human approves the best ones via a UI before video rendering begins

Actionable insights

→

Use a two-stage pipeline (image concept -> approval -> video) to validate ideas cheaply before spending credits on full video generation

→

Model parallel agent queues using Archon RLAIF loops: workers check a shared queue, claim one item, process it, then signal completion — making scaling to any catalog size trivial

→

Higgsfield CLI lets coding agents generate videos entirely from the terminal without leaving an agent session — enabling full automation without a browser UI

→

Install the Higgsfield CLI with two commands (documented in the repo readme) and pass branded product images as reference material to ensure logos and product appearance are accurate

→

The entire system (Archon workflows, catalog site, sample videos, Higgsfield skill) is available in a linked GitHub repo for immediate experimentation

Notable quotes

“We definitely can't have it try to generate a ton of videos all at once in a single coding agent session cuz it's going to get totally overwhelmed by that task.”

“A RLAIF loop is a harness — we can build any harness in Archon. We are building a RLAIF loop here for content. It is just that instead of our larger task list being a bunch of things to do in a codebase, it is the individual products we want to generate content for.”

“You can not really tell that is not a real person. I am just really impressed with what Higgsfield is able to do here.”

Worth watching?

✅

Worth watching the full video?

Watch if you want to see the live demo and workflow diagrams — the key architecture and steps are all captured here, but the actual generated UGC video quality is best judged by watching.

Topics

AI & Tech Claude Code

Explore more summaries on these topics →

Saved you some time? The creator still deserves a like.

Watch on YouTube →

More like this