Rourke Heath – Universal AI Cinematic Automation

Here’s What You Get:

• 18 environment reference images — ready to use
• Two years of prompt knowledge and frameworks for image and video generation
• The full 12-step INSTRUCTION-MANUAL.md
• Reference, logging, and failure-map templates
• The Google Sheets CSV shot tracker
• Content-safety guardrails and safe rewrites
• HANDOFF.md, README.md, and all working subfolders

It’s a whole production pipeline.

In one folder.

This doesn’t treat your AI agent as a one-off image generator. It turns it into a proper cinematic automation engine that handles the whole job, start to finish. Claude takes three roles at once.

Role 01

Prompt Engineer

Writes, refines, and fires the prompts for stills and multi-shot video. You give the direction in plain English — it handles the prompt craft.

Role 02

Production Manager

Runs the Higgsfield CLI, captures your reference UUIDs, sorts the output folders, and keeps the shot tracker up to date.

Role 03

Quality Controller

Reads what came back, logs the failures, applies safe rewrites, and keeps going until the shot’s approved.

Everything the studio runs on:

Environment library

18 stunning environment images

Eighteen ready-made environment reference images come in the pack — fully realised worlds you can drop straight into a shoot or pull apart for your own. Each ships with a matching text descriptor, so you can set a scene without the background bleeding in.

Environment by text

No ref bleeding

Control lighting, mood, and atmosphere with descriptors instead of image refs. It sidesteps the common ref-bleeding problem completely.

Prompt kit

Prompt frameworks

Ready-made structures for stills and 5 to 15 second cinematic video with timestamped cuts, camera moves, physical sound design, lighting, and genre flags.

Pipeline

The production pipeline

Automatic ref-ids.md, prompt-log.md, seedance-failures.md, batched output folders, and a Google Sheets CSV shot tracker.

Guardrails

Content-safety guardrails

Pre-documented NSFW and IP triggers, safe rewrites, ref-stacking limits, and resolution rules so generations do not fail silently.

Feedback loop

It learns your brand

Claude reads previous results and iterates. After a few campaigns you hold a personalised record of what works for your brand.