FlowMuse

A local-first workbench for AI images and video: prompts, generation tasks, project assets, conversational workflows and a sketch canvas in one place, also packaged as a desktop app.

Year: 2026
Stack: TypeScript · React · NestJS · Electron
Highlight: Open source · MIT

AWhat it is

FlowMuse is a local-first workbench for creating with AI. Image and video generation, reference material, project context and prompt refinement sit together, and everything it makes is saved on your own machine.

BWhat it does

Quick mode generates images and video from text or images; workflow mode drives an image agent, video shot planning and automatic creation through conversation; the canvas is for sketches and annotations; the task centre tracks every job’s status, failures and results. Channels such as NanoBanana, Midjourney, GPT Image, Doubao, Qwen and Wan are built in — add a base URL and an API key and they work.

CHow it’s built

React 19, Vite 6 and Tailwind CSS on the front, NestJS 10 on the back, with data in SQLite through Prisma. The same code runs as a browser service or, through Electron, as a desktop app for macOS and Windows; FFmpeg stitches planned shots into a video.

DWhere it is

Open source under the MIT licence; the code and setup instructions are on GitHub.

FlowMuse conversational creation — Conversational creation: plan shots in a chat and submit image and video tasks from it.

The FlowMuse canvas — The canvas: sketch and compose, then export a PNG or save it to the project.

EThe lines that matter

npm install && (cd frontend && npm install)cp .env.example .envnpm run prisma:generate && npm run prisma:initnpm run dev:all        # browser modenpm run desktop:dist   # macOS / Windows installers

GitHub ↗

003Next projectPlants vs. Zombies, in CA course project: Plants vs. Zombies written in C with the EasyX graphics library, and the whole build written up in fifteen posts.