Fogsight Lets an LLM Direct Animated Explainers From Text

What it does

Fogsight is an animation engine that takes abstract concepts—like “entropy increase” or “bubble sort”—and generates complete animated shorts with bilingual narration and what the project calls “cinematic” visual quality. Users type a prompt, and an LLM handles the rest: scripting, visual composition, and motion. A local web interface and a hosted version at fogsight.ai both let users iterate on results through a chat-like Language User Interface.

The interesting bit

Instead of treating the LLM as a mere script generator, Fogsight uses it as an orchestration layer that coordinates the full pipeline—narration, visual elements, and dynamic effects—like an AI director rather than a writer. The “Language User Interface” then lets you refine the output conversationally, which is a neat answer to the “prompt and pray” problem.

Key highlights

• Generates narrative-driven animations from single text prompts, with examples covering physics, math, CS, and design concepts.
• Runs locally via Python/Docker or works through the hosted fogsight.ai service.
• Uses an OpenAI-compatible SDK but explicitly recommends Google Gemini 2.5 Pro for the backend.
• Supports iterative refinement through a multi-round chat interface dubbed “Language User Interface” (LUI).
• Outputs include bilingual narration (Chinese and English in the showcased examples).

Caveats

• The project ships under a CC BY-NC-ND 4.0 license, meaning commercial use and derivative forks are prohibited without contacting the authors.
• The README is primarily in Chinese; English documentation exists but appears secondary.
• It is unclear from the README exactly which rendering engine or video format is used under the hood, or how long typical generation takes.

Verdict

Science communicators and educators who need quick visual explainers should look here; developers wanting a commercially reusable video engine should look elsewhere due to the restrictive license.