Stonewuu/ai-fusion-video

What it is

Rongguang is a self-hosted platform for AI-assisted video production, aimed at creators working on short dramas, motion comics, and similar content. It puts projects, scripts, storyboards, assets, and image/video generation into one workspace, so each shot can be tracked from script breakdown to final composition.

Its distinguishing pieces are an infinite canvas that links script, storyboard, character, and asset nodes, an in-app agent ("Rongguang Assistant") that can rewrite text, organize storyboards, or trigger generation using project context, and ComfyUI workflow integration. The agent can also use MCP-connected external tools or delegate subtasks to sub-agents; tool access is authorized per conversation.

Model API details are configured in the settings UI rather than hard-coded. Chat and agent features support OpenAI-compatible endpoints, Anthropic, Gemini, DashScope, and Ollama; image and video generation support multiple backends, subject to the provider account and model permissions. Media and agent workspaces can use local disk or S3-compatible storage.

Quick start

The recommended path is Docker Compose. Install Docker Engine and Docker Compose, then from the repository root:

git clone https://github.com/Stonewuu/ai-fusion-video.git
cd ai-fusion-video
cp .env.example .env
docker compose up -d

This starts MySQL, Redis, the backend, the frontend, and Nginx. Open http://localhost:8080 and follow the initialization wizard to create the administrator account. To check status or follow backend logs:

docker compose ps
docker compose logs -f backend

Custom ports, passwords, and deployment settings go in .env, modeled on .env.example. After startup, configure model endpoints and keys, then connect a ComfyUI service if you want to import API-format workflows and bind them to image or video models as described in the workflow section of the project README.

Source development

Running from source requires JDK 21, Node.js 20 with pnpm 10, Docker for MySQL and Redis, and FFmpeg/FFprobe for composition and media probing. Start middleware with docker compose -f docker-compose-middleware.yml up -d, then run ./mvnw spring-boot:run in the backend directory (on Windows, mvnw.cmd). In a second terminal, run pnpm install and pnpm dev in ai-fusion-video-web. The front end proxies /api/** and /media/** to port 18080; backend Swagger UI is available at http://localhost:18080/swagger-ui.html.

Deployment constraints

• Public deployments must change the default MySQL root and Redis passwords. The defaults are for local use only.
• In the combined-entry deployment, PUBLIC_API_URL must stay empty. It is only filled when frontend and backend use separate public domains.
• For separated deployment, PUBLIC_API_URL is the backend root address without /api, and CORS origins should be restricted to the real frontend origin in production instead of *.
• Cloud image/video models generally cannot read localhost or LAN reference assets. Use public object storage or expose the backend resource endpoint publicly, and save that public address in settings.
• The default Compose setup does not publish MySQL or Redis ports to the host. Host access requires uncommenting the relevant ports configuration.

HotRepo's take

The project appeared on the Java daily list on 2026-11-05 with a best position of 9, at 1,711 stars and 10 stars gained that day. That is a single snapshot rather than a trend, but it shows the repository drew attention within the Java segment on that date. The README leads with sponsor banners and QR codes before the product description, which is worth knowing when reading the page, but the repository itself documents a substantial Java/Next.js implementation, deployment options, and a source-development path. For readers evaluating it, the more useful signal is the concrete stack and configuration detail rather than the sponsorship section.

Project source and documentation