manim-video

Write a plan with the narrative arc and scene list, then create one Manim Scene class per scene in a single Python script. Render drafts with manim -ql, stitch clips with ffmpeg concat, and render the final at -qh for 1080p60 output.

Use MathTex with raw strings for each step and TransformMatchingTex to morph between equations. For complex expressions, use substrings_to_isolate and key_map so individual terms match and animate correctly.

You need Python 3.10+, Manim CE v0.20+ installed via pip install manim, a LaTeX distribution such as texlive-full or MacTeX, and ffmpeg for video stitching. Run scripts/setup.sh to verify all dependencies.

Text objects are Mobjects, not VMobjects, so VGroup rejects them on Manim CE v0.20+. Use Group for mixed collections containing Text, and reserve VGroup for shapes or MathTex only.

Yes, use the manim-voiceover plugin with ElevenLabs, Azure, or Google TTS services to auto-sync animation duration to narration. Alternatively, mux audio manually with ffmpeg after stitching the scene clips.

Avoid Manim for single static labeled diagrams, dense reference tables, or concepts already clear from one figure, since motion adds cognitive load. It is also unsuited for text-heavy scenes rendered at low draft quality.