AnimatedVisuals

Convert complex, hard-to-understand concepts into smooth, narrated, animated videos — as if a world-class professor were teaching you.

Comment 1

Inspiration

I wanted to solve a problem I’ve faced countless times: how to make very hard concepts easy to understand. Reading textbooks often feels abstract, and watching plain lectures can be dry. But when a world-class professor explains with visuals, everything just clicks.

I’ve always admired creators like 3Blue1Brown, who make math and science look magical through animations. My inspiration was to automate that magic - to take any explanation and instantly turn it into a narrated, animated lecture video.

What it does

AnimatedVisuals converts raw explanations into segmented, narrated, and animated lecture videos.

  1. Segments the input into 2–6 smooth, professor-like lecture units.
  2. Generates natural narration with pauses, emphasis, and realistic pacing.
  3. Produces Manim-based animations that illustrate the concepts step by step.
  4. Syncs audio and visuals so the video feels seamless.
  5. Concatenates everything into one final MP4, ready to share with students.

The output feels like you’re learning from a top educator, not a machine.

How I built it

  • Segmentation: I used Groq LLMs to break down explanations into 2–6 smooth segments with narration scripts and visual descriptions.
  • Narration: I integrated Azure Speech Services to generate lifelike audio, complete with pauses and emphasis.
  • Animation: I converted visual descriptions into Manim (v0.19) scripts, with retries and repair logic to keep the code error-free.
  • Syncing: I used ffmpeg and MoviePy to align narration with animations and handle duration mismatches.
  • Assembly: I concatenated all the segments into a final lecture video with smooth transitions.
  • CLI/UX: I built a terminal-first pipeline with rich for progress feedback, and an easy .env-based configuration.

Challenges I ran into

  • Manim complexity: Generating correct animations from natural language was tricky. I built a repair loop that retries and fixes broken code automatically.
  • Audio-video sync: Narration and visuals often ran at different speeds. I solved this by measuring audio duration and adjusting self.wait and run_time.
  • Consistency across segments: Ensuring transitions felt smooth (not jumpy) required prompt engineering and duration tuning.
  • Time: Building a full end-to-end pipeline in a hackathon window meant focusing on core templates first, then extending.

What I learned

  • How to design LLM prompts that output structured JSON for both narration and visuals.
  • The power of segment-based pipelines: processing audio + visuals per segment makes debugging and syncing easier.
  • How to integrate multiple AI services (LLM + TTS + animation) into one seamless tool.
  • That even advanced math concepts can feel intuitive when explained with the right timing and visuals.

Accomplishments I’m proud of

  • A working prototype that can turn something as abstract as “derivatives” into a professor-like animated mini-lecture.
  • Smooth audio narration with realistic pauses and pacing.
  • Error-repairing Manim code generation, no manual edits required.
  • Building a foundation that could scale to K–12 education and beyond.

What’s next

  • Adding a UI interface for teachers and students (beyond CLI).
  • Expanding visual templates (e.g., physics diagrams, chemistry animations).
  • Support for multi-language narration.
  • Integration with learning platforms (Google Classroom, Canvas, etc.).
  • Enabling teachers to edit or refine segments before final render.

Built With

Share this project:

Updates