ayushnangia/paper-to-animation

How to use

Read individual rule files for detailed explanations and code examples:

Pipeline Phases

• rules/paper-comprehension.md - Phase 1: Read the paper, find the story arc, extract headline numbers
• rules/storyboard-design.md - Phase 2: Design 5-6 scenes with archetypes and timing
• rules/data-extraction.md - Phase 3: Pull exact numbers into Python data structures with colors
• rules/character-design.md - Phase 4 (optional): Build a persistent character from Manim primitives
• rules/manim-script.md - Phase 5: Generate the Manim script with scene methods and patterns
• rules/rendering.md - Phase 6: Render, export GIF, iterate

Manim Techniques

• rules/text-rendering.md - Width-aware Pango text helper, multi-line patterns, DecimalNumber pitfalls
• rules/color-palettes.md - Dark/light themes, AI company brand colors, chart differentiation
• rules/scene-transitions.md - Scene wipe patterns, persistent elements, transition timing
• rules/common-mistakes.md - Common pitfalls and their fixes

Working Examples

Complete worked example demonstrating the full pipeline:

• examples/deep-thinking-tokens/storyboard.md - 6-scene storyboard for "Think Deep, Not Just Long" (arXiv 2602.13517)
• examples/deep-thinking-tokens/deep_thinking_video.py - Full Manim script (626 lines)
• examples/deep-thinking-tokens/deep_thinking_video.gif - Final output

Templates & References

Copy and modify these to start new projects:

• references/storyboard-template.md - Blank storyboard with scene archetypes
• references/manim-scaffold.py - Starter Scene class with txt(), scene_wipe(), pill(), glow_highlight()
• references/ffmpeg-recipes.md - Complete ffmpeg commands for GIF, clips, thumbnails, scaling

Quick Reference

Checklist

IMPORTANT: Use TaskCreate to create a task for EACH phase below.

• [ ] Phase 1: Paper Comprehension
• [ ] Phase 2: Storyboard Generation
• [ ] Phase 3: Data Extraction
• [ ] Phase 4: Character Design (optional)
• [ ] Phase 5: Manim Script Generation
• [ ] Phase 6: Rendering & Export

Core Principle

Narrative first, code second. The storyboard drives everything.

Announce at start: "I'm using the paper-to-animation skill to build this video."

Script Structure

from manim import *

class PaperVideo(Scene):
    def construct(self):
        self.camera.background_color = "#0d1117"
        # Scene sequence
        self.scene_1_hook()
        self.scene_wipe(target_pos=UL * 2.5)
        self.scene_2_setup()
        self.scene_wipe(target_pos=UR * 2.5)
        self.scene_3_results()
        self.scene_wipe(target_pos=DR * 2.5)
        self.scene_4_insight()
        self.scene_wipe_simple(target_pos=LEFT * 3.5)
        self.scene_5_closing()

Render Commands

manim -ql script.py ClassName    # Test (480p, fast)
manim -qh script.py ClassName    # Final (1080p)

GIF Export

ffmpeg -i <mp4> -vf "fps=15,scale=720:-1:flags=lanczos,split[s0][s1]; \
  [s0]palettegen=max_colors=196[p];[s1][p]paletteuse=dither=bayer" \
  -loop 0 output.gif

Timing Guidelines

• Individual animations: 0.2-0.8s (keep snappy)
• LaggedStart lag_ratio: 0.07-0.15
• Scene holds for reading: 0.3-0.8s via self.wait()
• Scene wipe transitions: 0.4-0.6s
• Total: 25-35s for social media, 60-90s for full explainer

Key Differences from ManimGL

| Feature | Manim Community (use this) | ManimGL (3b1b) | |---------|---------------------------|----------------| | Import | from manim import * | from manimlib import * | | CLI | manim | manimgl | | Creation | Create() | ShowCreation() | | Text | Text(), MathTex() | Text(), Tex() | | Package | pip install manim | pip install manimgl |

Installation

pip install manim
manim checkhealth   # verify installation

Requires: Python 3.10+, ffmpeg, LaTeX (for MathTex)