An AI video factory that still needs a human foreman

What it does

融光 (Rongguang / “Fusion Light”) is a full-stack web app for producing AI short dramas and comic-style videos. You write a script; the system breaks it into storyboards via LLM, generates reference images through multiple AI art engines, then feeds those into video generators. The result is a browser-based production pipeline that keeps humans in the loop at the script and review stages.

The interesting bit

The “Agent” framing is mostly workflow orchestration — Spring Boot controllers calling LLM APIs in sequence — rather than autonomous agents making creative decisions. The actual value is the plumbing: it unifies OpenAI, Claude, Gemini, Qwen, DeepSeek, and Ollama under one configuration panel, and handles the tedious asset handoff between text → image → video stages that currently breaks most manual workflows.

Key highlights

• Java 21 / Spring Boot 3.5 backend with Spring AI abstraction layer; Next.js 16 frontend
• Multi-model LLM support with runtime provider switching via system settings
• S3-compatible storage backends (OSS, COS, MinIO, local) configurable without restarts
• Visual pipeline view for tracking where a project sits in the generation queue
• Docker-first deployment; claims one-command startup

Caveats

• Team collaboration and permissions are on the TODO list; currently single-user or shared-login only
• “Agent” is aspirational — the roadmap lists a “global intelligent Agent” as unfinished
• Heavy reliance on external API providers; no local inference stack beyond Ollama support

Verdict

Worth a look if you’re building content farms or short-drama studios and want to stop juggling five browser tabs and a spreadsheet. Skip it if you need real multi-user collaboration today, or if you were hoping for genuinely autonomous creative agents rather than a well-organized API router.