From Abstract Words to Cinematic Animation in One Click: A Complete Guide to Running Fogsight Locally

1. Why Fogsight Matters Today

Problem Traditional Workflow How Fogsight Changes the Game
Explaining a concept 2–3 days of slide decks, screen recordings, storyboards Type a word, wait 10–20 min, get a bilingual narrated short film
Iterating on feedback Re-render, re-record, version hell Chat with AI in plain English, see changes in real time
Required skills After Effects, Premiere, scripting Laptop + browser is all you need

In short, Fogsight (雾象) turns prompt-writing into directing, letting anyone create cinema-grade micro-movies from a single sentence.

2. What Can Fogsight Actually Do?

2.1 Core Use Cases

  • Teachers: turn “Euler’s formula” into a 30-second clip for MOOCs.
  • Science communicators: explain the second law of thermodynamics with an “entropy increase” animation that hit 100 k views on Bilibili.
  • Product managers: demo “affordance in design” to stakeholders in 60 seconds instead of a 30-slide deck.

2.2 Public Examples (click to watch)

Input Phrase Discipline Length Watch
entropy increase Physics 50 s Bilibili BV1PXgKzBEyN
Euler’s formula Mathematics 45 s Bilibili BV1yXgKzqE42
bubble sort Computer Science 60 s Bilibili BV1sQgKzMEox
affordance in design Design 55 s Bilibili BV1yQgKzMEo6

All clips include bilingual narration and are ready to embed in classrooms or social media.

3. How It Works (High-Level)

  1. Large Language Model (LLM) parses your prompt and decomposes it into “shot list + narration.”
  2. Animation Orchestration Engine maps the list to visual assets, transitions, and sound.
  3. Language User Interface (LUI) lets you refine the result conversationally: “make the second scene warmer” updates everything in seconds.

4. Local Installation Walk-Through

Verified on macOS 14, Ubuntu 22, Windows 11.

4.1 Prerequisites

  • Python 3.9+
  • Modern browser (Chrome ≥ 120, Firefox ≥ 121, Edge ≥ 120)
  • LLM API key (Gemini 2.5 Pro recommended)

4.2 Step-by-Step Commands

# 1. Clone the repo
git clone https://github.com/fogsightai/fogsight.git
cd fogsight

# 2. Install dependencies
pip install -r requirements.txt

# 3. Configure credentials
cp demo-credentials.json credentials.json
# Edit credentials.json
#   "API_KEY": "your_Gemini_2.5_key",
#   "BASE_URL": "https://generativelanguage.googleapis.com/v1beta/openai/"

# 4. Launch
python start_fogsight.py

Your browser will open automatically at http://127.0.0.1:8000.

5. Your First Animation: “Bubble Sort” in Practice

  1. Type bubble sort in the input box.
  2. Click Generate.

    • Progress bar: script → storyboard → voice-over → final render
    • 720p ~ 5–8 min, 1080p ~ 10–15 min on RTX 3060
  3. Preview the clip.
  4. Chat refinement:

    • “Slow down the third pass by 0.5 s and change the background to light blue.”
    • New version ready in ~30 s.

6. Frequently Asked Questions

Question Straight Answer
Does it support Chinese? Yes—voice-over in Chinese, English, or bilingual.
Why is the first run slow? Initial model download; subsequent runs use cache and are 3–4× faster.
Can I run it offline? LLM still needs internet; rendering and caching are local.
Output format? MP4 (H.264), 720 p / 1080 p, 8 Mbps.
Commercial use? MIT license—commercial use allowed; attribution appreciated.
No GPU? CPU-only works; 720 p takes ~15 min. RTX 3060+ cuts time by 4–5×.
How to join the community? Visit the Feishu doc and click “Join Group.”

7. Pro Tips for Better Clips

7.1 Prompt Tricks

  • Add domain keywords:
    Euler’s formula mathematics polyhedron → more accurate shots.
  • Specify length:
    30-second short vs 90-second detailed version.
  • Language hint:
    bilingual narration or English only.

7.2 Ready-to-Copy LUI Commands

Goal Command
Change color scheme “Switch to cyber-punk palette, main colors #00ffff and #ff00ff.”
Add subtitles “Add white subtitles with black stroke, 48 pt, bottom center.”
Adjust pacing “Extend the second scene by 2 s and add fade-in/out.”
Swap voice “Use gentle female voice, speed 0.8×.”

8. Contributors & Credits

Fogsight is part of the WaytoAGI open-source initiative.

  • Universities

    • Taited – PhD student, CUHK-Shenzhen
    • yjydya – PhD student, NTU

  • WaytoAGI Community

    • richkatchen (陈财猫)
    • kk

  • Index Future Lab

    • Lin He

  • AI Explorers & Independent Devs

    • Xiaodao Huang, shuyan-5200, Ruyue Wang, Jack-the-Builder, xiayurain95, Lixin Cai

Thank you to every bug-reporter and feedback-giver in the Feishu group.

9. Roadmap

  • Plugin system: upload custom visual assets.
  • Local fine-tuning script: run 7 B LLM on 8 GB VRAM for offline generation.
  • Community template library: one-click reuse of “Fermat’s Little Theorem,” “Quicksort,” etc.

10. Take Action Today

You now know:

  • Fogsight turns abstract terms into animated explainers.
  • Local setup is five commands away.
  • Iteration is as simple as chatting with AI.

Open your terminal and run:

git clone https://github.com/fogsightai/fogsight.git

Ten minutes from now, you could have a 45-second short on Schrödinger’s cat. Share your first clip in the Feishu group and let’s make knowledge visible together.