Create narrated Manim animations in one Windows workspace. Chat with Codex or Claude Code to generate the scene, narration and assets, or edit Python directly. Render and preview videos without installing Python.
Download Manim Studio · Report a bug
1.4.0 makes narration sound natural by timing the animation to the voice, cleans up generated videos, and adds redesigned house styles set in bundled Google Fonts, with live style previews, per-project fonts and a Queue button.
Requires 64-bit Windows 10 (2004+) or Windows 11. Allow 3 GB of free space during setup. The app is unsigned, so Windows may display a security warning.
| Download | Use it when |
|---|---|
| Setup | You have internet. Setup downloads and prepares the rendering engine. |
| Portable ZIP | You want a folder you can move. Extract the entire ZIP, then open Manim Studio.exe. |
| Offline setup | You need to install without downloading rendering tools. |
All downloads include standard LaTeX and Typst equation support. Studio uses shared Microsoft WebView2; portable first launch needs internet only if that runtime is missing. Standard rendering works offline after setup.
First launch opens a setup guide that checks for Codex and Claude Code, offers installation and login, and explains the workspace. You can skip setup to edit scripts manually, or reopen it from Settings → General → Open setup.
- Pick a project in the sidebar, or select New project to start one. Each project keeps its own script, narration and assets, and saves automatically.
- Describe your video in chat, or paste a complete Manim Community script in Script.
- In Preview, choose an output preset. Custom sets exact dimensions and frame rate. Studio selects the animation automatically; Animation appears only when the script contains several.
- Click Render or press Ctrl+Enter. Agent changes render automatically when enabled in Settings.
- Preview the result and save the MP4. Find previous work, and the queue of videos waiting to render, in Renders & queue (or the Queue button beside Render).
The editor includes autocomplete, formatting, and errors linked to script lines. Script errors after generation or rendering go back to the same chat automatically. Studio retries rendering after each repair, up to three repairs, and shows progress in chat, Preview and Renders & queue. Stop video cancels the remaining work. You can also choose Send to LLM beside a script error or failed render to start repair yourself. Provider/login failures or repeated unsuccessful repairs pause the video with details. Scripts and agent commands run with your Windows account's permissions; review code before rendering.
Open Settings → Providers. Studio detects locally installed CLIs, shows their version and login status, and provides the install command and Log in action. The first-run setup can open each CLI's installer in PowerShell. Codex's npm installation requires Node.js LTS; setup links to its download page. Complete login in the CLI window, then choose Check again. You can override the binary path, choose a default model and reasoning effort, and add custom model IDs. Codex's model list comes from the CLI's live catalog.
Each chat belongs to a project and resumes its provider session on later messages. The timeline streams replies, commands, file changes and errors. Stop cancels the process tree. Start a new chat to change providers after a session is established. Your existing CLI login stays with the provider; Studio does not store credentials. Agent generation requires the CLI's account access and internet. The rendering engine and downloaded voice models work offline.
Full access allows file edits and commands without CLI permission prompts. Auto edits uses the Codex workspace sandbox or Claude's acceptEdits permissions; Claude's mode is not an operating-system sandbox. Agents are instructed to use the installed libraries, create local assets and run Studio's scene checks.
Click a project folder to expand or collapse its chats. Studio remembers your choices.
Collapse the sidebar into an icon rail with the button beside Manim Studio. The rail keeps New project, an import/PDF menu and a searchable project switcher at the top, with Settings, Renders & queue, usage and updates below. Settings and Renders & queue share a Back button in the sidebar footer; on phones it appears in the page header. Use the paperclip in chat to upload reference documents, images or audio. Uploads finish before the message can be sent. The ⋯ button on each row opens Rename, Open project folder and Delete, including for projects you are not currently editing.
Deleting asks for confirmation and removes that project's script, narration and assets. Finished videos remain in Renders & queue. You cannot delete a project while it is rendering. Studio starts with no project folders until you create or import one. Deleting the last project returns to the empty workspace.
Current release: 1.4.0. See the release notes.
Choose Videos from PDFs in the sidebar to start several videos together.
Add one script PDF per video, choose an assistant, and click Create and start videos.
Choose Same folder for one new parent project, Separate folders for a
project per PDF, or Existing folder to add videos to a saved project. Each
PDF gets its own chat, script, narration and assets. Shared projects keep videos
in separate workspaces under their videos folder, so their files stay independent.
The original PDF and an extracted text copy are kept with each video.
PDFs need selectable text and can be up to 20 MB each.
Download the narration voice before starting if you want narrated renders.
With automatic rendering enabled in Settings, finished scripts enter the render queue. You can also click Add to queue while another video renders. Studio renders videos one at a time. Renders & queue shows the queue: pause or start it, move waiting videos up or down, clear them, or cancel them individually. Two queue preferences (here and in Settings → General) control how it behaves:
- Start is Automatically by default. Choose When I start the queue to hold new videos until you press Start queue; the queue returns to holding once it has finished everything.
- If a video fails is Fix and re-render by default: the agent repairs the script in the same chat (up to three attempts) and the video rejoins the queue while the other videos keep rendering. Skip it leaves the failed video in Renders & queue and carries on; Pause the queue stops after the failure.
Keep Studio open until the queue finishes; waiting jobs are cancelled when Studio closes.
The folder selector and project search filter the sidebar. The Usage limits button beside Settings shows the remaining ChatGPT/Codex and Claude plan limits reported by your signed-in CLIs.
Open the Assets tab to add images, SVGs and sounds to the current project.
Refer to them in your script as assets/<file name>. Copy the path beside an asset,
or attach reference files to your chat and ask the agent to use them.
Select Import project in the sidebar, then choose a project folder or ZIP. Studio creates a separate project, loads the scene and narration, and keeps nested asset paths intact. Select Copy project prompt for instructions you can give an online AI that creates downloadable files. Review the imported code, download the voice model if prompted, and select Render.
The minimal package is:
My animation/
script.py
narration.txt # optional timed voice-over
assets/ # only referenced images, fonts or audio
Other scene filenames work when the folder contains one scene script. For several
scene files, include project.json with {"name":"My animation","entrypoint":"main.py"}.
Studio also reads narration.json as a list of {"start":0,"text":"Hello"}
segments, or an object with segments, voice, speed and quality settings.
Existing TSkill chapter lists are supported. Generated previews, installation
helpers and reports are omitted; imported scripts never install dependencies.
Packages can contain up to 2000 files and 512 MB of unpacked content.
The Narration tab adds a spoken voice-over to the finished video. Speech is generated on your computer with the Kokoro-82M voice model, which runs on the CPU without a GPU. Speed depends on your computer. The first use downloads the model once (about 350 MB for High quality, the default, or 120 MB for Standard); after that narration works offline.
Write a few flowing paragraphs, one per idea. Paragraphs that follow each other
are spoken in one go, so they keep a natural rhythm. Start a paragraph with a time
such as [0:05] to begin speaking at that moment; otherwise it follows the
previous line. Lines starting with # are notes. Narration is mixed with any sound
your scene already adds. If the narration is longer than the animation, the last
frame holds; add self.wait() at the end of the scene to plan for it.
To make the animation wait for the voice instead, import Studio's helper and wrap the animations for each paragraph. The block lasts as long as that paragraph takes to speak, measured from the real voice when you render:
from studio_narration import narrate
with narrate(self, 1) as line: # the first paragraph of narration.txt
self.play(Create(graph), run_time=line.duration / 2)The chat agent writes scenes this way. For a more natural delivery, write connected sentences with punctuation and spell equations out in words. Studio normalizes notation, adds sentence pauses, softens abrupt audio boundaries and levels loudness. Try different voices and speeds with Preview voice; High quality uses the full-precision model. Naturalness still depends on the selected voice and wording.
Installed copies save work in %LOCALAPPDATA%\ManimStudio\Data. Portable copies
use UserData beside the executable. Projects (scripts, narration and assets) are
in its projects folder; render history, preferences and the voice model stay in
the same data folder. Use Settings → About & updates to check for future releases.
Each project stores script.py, narration.txt, voice settings in project.json
and its own assets/ directory. The project's ⋯ → Open folder action reveals
these files.
Chat sessions and provider preferences stay in .studio. Project edits save
automatically; failed saves show a retry action, and recoverable drafts survive a
reload. The desktop window saves pending changes before it closes.
With Python 3.12, uv, and Node.js installed:
uv sync --locked
npm ci
npm run build:editor
uv run python desktop.pySee AGENTS.md for checks and runtime packaging for build profiles and equation compatibility.
Created by ItsTatsuya. App code is MIT licensed. Built on Manim Community and FFmpeg. Third-party licenses remain with their components; matching sources and checksums are included with each release.