manim-narration

name

manim-narration

description

Trigger when: (1) User asks for a "narrated Manim video", "anlatımlı manim", "manim with voice", "manim video with audio narration", or "PR walkthrough video"; (2) Code imports from manim_narration.lib.* or uses NarratedScene; (3) User wants Manim output that includes spoken narration synced to animations. Builds narrated Manim videos using Microsoft edge-tts (free, no API key) for voice synthesis and ffmpeg for proper audio re-encoding so MP4s play reliably in VSCode, Chromium, and Electron previews. Includes a NarratedScene base class that exposes a `narrate(text)` context manager paced to real audio duration. Optionally uses gstack /browse to pre-fetch live screenshots declared via `# ASSET:` manifest comments — so the video can showcase real websites, GitHub PRs, dashboards, IDE views, etc., without manual capture. Composes with manimce-best-practices for animation idioms. NOT for ManimGL — use manimgl-best-practices there (no narration support). Dependencies: installed automatically by ~/.claude/install.sh. - pip: manim, edge-tts - system: ffmpeg, ffprobe - optional: gstack (already installed by ~/.claude/install.sh)

Manim Narration

A 30–120s narrated MP4 with synchronized voice, cinematic transitions, and optional live-website screenshots — playable in any modern player including VSCode's preview pane.

When to invoke

• "make a narrated manim video about X"
• "explain Y in a 60s manim animation with voice"
• "PR walkthrough video for "
• "anlatımlı manim videosu yap"

Quick usage

1. Subclass manim_narration.lib.voiceover_scene.NarratedScene.
2. Wrap each beat: with self.narrate("Sentence here.") as t: ...
3. (Optional) Add # ASSET: <slug> | <url> | viewport=WxH | clip=<css> manifest comments. Run the visuals pre-fetch before rendering.
4. Render with manim -qh, then re-encode audio with ffmpeg per RENDER.md.

Override voice per scene

class MyScene(NarratedScene):
    VOICE = "tr-TR-EmelNeural"   # Turkish female

Tested voices: en-US-AriaNeural (default), en-US-JennyNeural, en-US-GuyNeural, tr-TR-EmelNeural, tr-TR-AhmetNeural. Avoid en-US-AvaNeural (flaky).

Visual asset pipeline (gstack /browse)

Pre-fetches screenshots from real URLs before render so scenes can ImageMobject(asset("slug")) real web content. Falls back to a manual capture instruction list if gstack is unavailable. See RENDER.md.