PAI Pro Just Put an Entire AI Film Studio Inside Your Coding Agent

AI Filmmaking Just Left the Browser and Moved Into Your Terminal

Utopai Studios just open-sourced PAI Pro, the AI filmmaking engine that powered “Chloe vs. History,” the AI series that pulled over 600,000 followers and millions of views. And it does not run in a web app. It runs inside your coding agent.

Think of it this way: most AI video tools give you a prompt box and a render button. PAI Pro gives you a canvas, a timeline, a script engine, and character persistence across every single shot. And you drive it all from Claude Code, Cursor, or Codex. No tab switching. No separate platform.

What PAI Pro Actually Does

PAI Pro is not another “type a prompt, get a clip” tool. It is filmmaking infrastructure built as agent skills. Here is what that means in practice:

• Character persistence. Define a character once. The same face, wardrobe, and proportions show up in every image and every video clip across your entire project. No more rolling the dice on whether your AI character looks the same in shot 2 as shot 1.
• Script-to-shot orchestration. PAI Pro ingests your screenplay, breaks it into shot-sized beats, groups them into scenes and acts, and tracks everything as a directed graph. When you say “shot 3” or “the morgue scene,” it knows exactly what you mean.
• A React Flow canvas. Visual node editor where you lay out characters, locations, images, video clips, and notes. Group them into scenes. See your entire production at once.
• A timeline. Drag clips onto a reel, reorder them, scrub through your sequence. This is not a folder of clips. It is a cut.
• Director-level control. Specify camera angles, framing, lens choices, and shot composition. The agent translates your direction into provider-specific generation cues.
• Per-project memory. Every project owns its own workflow file and asset folder. Switch projects and the agent context follows.

How It Works: Skills, Not APIs

PAI Pro is delivered as installable skills in standard SKILL.md format. If you use Claude Code, Cursor, or Codex, you already know how skills work. Install them and your agent gains new capabilities without leaving your terminal.

The architecture is a hierarchy, not a flat set of endpoints:

• The orchestration layer owns the canvas, the script, and the structure of your project. It tracks every node, every scene, every act. This is what turns a folder of generated clips into a coherent film.
• The asset layer is a set of narrow, single-purpose skills. One generates concept art. Another generates character keyframes. Another creates motion clips. Another handles voice and narration. Each does one thing well, and the orchestrator decides how they compose.

This is the distinction that matters. A tool generates a clip. An infrastructure understands how that clip fits into a scene, how that scene fits into a sequence, and how that sequence serves a story.

Getting Started

PAI Pro is open source and local-first. Everything runs on your machine. Generated media mirrors to your project folder. The only thing that leaves your machine is the actual generation calls.

Here is the quick setup:

1. Clone the repo: git clone https://github.com/Utopai-Research/pai-pro.git ~/pai-pro
2. Set up with Docker: cd ~/pai-pro && cp .env.example .env then add your PAI_KEY
3. Run: docker compose up --build
4. Open the app and run /login in the embedded terminal

Or run it directly without Docker:

1. git clone https://github.com/Utopai-Research/pai-pro.git ~/pai-pro && cd ~/pai-pro
2. ./setup && npm --prefix server install && npm --prefix web install
3. cp .env.example .env then add your PAI_KEY
4. ./start.sh

You get a PAI_KEY from the developer platform at pai-pro.utopaistudios.com. One key drives every capability: image, video, voice, and reference-asset uploads.

What This Unlocks

Right now, most AI video generation is a single-shot game. You prompt, you get a clip, you hope it looks good. If you need a second clip with the same character, you roll the dice again. If you need a sequence of 10 clips that tell a story, you are in for a long night of prompt engineering and manual consistency checks.

PAI Pro changes that equation because it was built as production infrastructure, not a consumer tool. The team behind it comes from Google, Meta, Adobe, and Princeton. They built it to produce actual series content at scale. “Chloe vs. History” is not a demo. It is a shipped product with a real audience.

What to look out for:

• Character-consistent storytelling. If you have been frustrated by AI characters that shift between shots, PAI Pro’s persistence system is the fix. Define once, maintain everywhere.
• Agent-driven filmmaking. The fact that PAI Pro runs inside your coding agent means you can script, iterate, and refine without ever leaving your workflow. No copy-paste between browser tabs.
• Open-source, self-hosted. Your project files and generated media live on your disk. No platform lock-in. The skill format is extensible, so you can write your own filmmaking skills.
• The skill ecosystem will grow. Right now PAI Pro covers image, video, voice, script, and timeline. Sound design, color grading, and score are coming as new asset generators under the same orchestration layer.

The gap between “prompt a single AI clip” and “produce a coherent AI film” has been the biggest blocker in AI video. PAI Pro is the first tool that actually bridges it, and it does it by meeting you where you already work: your terminal.

Want more like this? Join MOKU Club for free. Weekly resources, early access to new guides, and occasional templates you can actually use. Join below.