Utopai-Research/pai-code

PAI‑Pro – a local‑first AI filmmaking studio

What it is – PAI‑Pro is an open‑source workstation that lets you create short films (script, images, video, voice, up‑scaled 4K) using large‑language‑model agents (Claude Code or OpenAI Codex). The agents run locally, have persistent context tied to your project files, and expose a single API that covers all media‑generation steps.

Key components

• Agent integration – Connect a Claude Code or Codex‑powered agent to your project folder. The agent can read/write files, remember prior steps, and invoke specialised “filmmaking skills”.
• End‑to‑end media skills – Built‑in pipelines for:
  • script writing
  • image generation (standard and “pro” quality)
  • video generation (from generated images and audio)
  • voice synthesis
  • 4K up‑scaling of video
• Visual canvas & timeline – A web UI where you arrange scenes, manage assets, and see a timeline of shots, similar to a lightweight video‑editing board.
• Unified API service – One endpoint (PAI_KEY) that routes calls to the appropriate skill (image, video, voice, upscaler). This avoids juggling separate provider keys for each media type.

Quick start

1. Clone the repo

   git clone https://github.com/Utopai-Research/pai-pro
   cd pai-pro
   

2. Choose an agent – either the default OpenAI Codex or Anthropic Claude Code.
   • Docker (recommended)

     # Codex (default)
     ./scripts/docker-start.sh
     
     # Claude Code
     PAI_DEFAULT_AGENT_ID=claude ./scripts/docker-start.sh
     

   • Local mode (if you prefer not to use Docker)

     ./scripts/stop.sh   # free ports if needed
     ./scripts/setup      # add --agent claude for Claude
     ./scripts/start.sh
     

3. Open the UI at http://localhost:7588 (Docker) or http://localhost:7443 (local).
4. Follow the on‑screen prompts to provide a PAI_KEY (generated on the PAI‑Pro developer platform) and start a new project.

API service snapshot

Endpoint	Typical quality	Latency	Cost (approx.)
generate_image	Great	10‑30 s per 16 images	$0.11‑$0.23 for 1K‑4K resolution
generate_image_pro	Best	3‑6 min for 32 images	$0.26‑$0.77
generate_video	Best	3‑6 min (creates 9 images, 3 video clips, 3 audio tracks)	$0.11‑$0.57 per second of video
upscaler	Best	~5‑8 min for a 1‑min clip	$1.35‑$2.35 for 1080p→4K
generate_voice	Good	5‑15 s	$0.01 per 500 characters

Resources

• Discord – community chat for help and showcase: https://discord.gg/hb4MkHND
• Docs – detailed setup, skill reference, architecture, and FAQ are in the docs/ folder.
• Contributing – see CONTRIBUTING.md; commercial use of the core skills requires a separate enterprise agreement.

License – PAI‑Pro is released under the PAI PRO Sustainable Use License, allowing personal, non‑commercial research and internal business use. Commercial exploitation of the built‑in skills or enterprise‑designated code requires a paid license.

---

PAI‑Pro lets developers harness LLM agents as creative collaborators for full‑stack film production, all from a single locally hosted service.