aqm857886159/Nomi

Open-source, local-first desktop app for AI video creation: write a script → generate images & video → edit on a timeline → export. Bring your own model & API key — everything runs on your machine. Built with Electron + React.

What it solves

Nomi addresses the difficulty of precisely communicating visual intent to AI generators. Instead of relying solely on text prompts and "guessing" with the model, it provides a professional director's workspace to control composition, character identity, and camera movement through 3D staging and structured workflows.

How it works

Nomi integrates the entire production pipeline into a single local-first application:

  1. Scripting & Storyboarding: Users write a story, and AI breaks it down into a detailed shot list (shots, characters, and prompts).
  2. 3D Director's Stage: Users can pose 3D mannequins, set camera angles, and define frames to create visual references that lock spatial relationships across shots.
  3. Canvas Generation: Images and videos are generated in parallel on a canvas, where users can perform in-app edits (background removal, element adjustment) and reuse assets across nodes.
  4. Timeline Editing: Generated clips are dragged onto a timeline for final assembly and export as MP4.
  5. Model Integration: It connects to various image and video models (e.g., Sora 2, Veo 3.1, Kling, Wan) via API keys or official CLI integrations (like Jimeng).

Who it’s for

It is designed for AI creators, directors, and storyboard artists who need granular control over their visual output and want a unified workspace to manage the process from script to final video without switching between multiple web tools.

Highlights

  • 3D Composition Control: Use a 3D stage to "build" the shot, ensuring consistent framing and posing.
  • Character Consistency: Project-level asset locking to prevent "face drift" across different clips.
  • Integrated Pipeline: A seamless flow from script $\rightarrow$ storyboard $\rightarrow$ canvas $\rightarrow$ timeline $\rightarrow$ export.
  • Headless Capability Core: An MCP (Model Context Protocol) implementation that allows programming assistants like Claude Code or Cursor to control Nomi via dialogue.
  • Local-First: Projects and assets are stored locally on the user's machine.