KinoDraw

Turn any script into a hand-drawn whiteboard explainer video, on your own computer. Free and open source, no watermark (just a 2-second end card you can turn off), works offline on Windows, macOS and Linux.

KinoDraw was called Doodle Studio before version 0.2.0.

Paste a script and KinoDraw produces a finished MP4. A drawing hand sketches cartoon doodles, handwritten notes and small charts while a natural voice narrates. Captions, chapters, music, a thumbnail and a transcript come with it. English and Chinese (中文) are both first-class.

• Works offline and for free. Voice, drawing and video are all made locally. No account and no API key needed. (You can also point it at your own voice server.)
• 1,760+ doodles. It ships 279 original drawings in one bold outlined style (plus a recurring narrator character) and 1,483 Microsoft Fluent Emoji redrawn to match.
• It explains, not just decorates. Numbers become big stats and 100-square grids, dates become timelines, "X is called Y" becomes a sticky-note definition, and quotes become quote cards. Every section ends with a takeaway note that pins onto the agenda.
• Everything is editable. Swap any doodle, fix a label, retitle a section, preview a frame, then make the video.
• Optional AI director. It can plan the visuals with GPT-6 Luna or Claude: through KinoDraw Cloud (free, no account or key needed), or with your own OpenAI, Anthropic or OpenAI-compatible key.
• Style: Choose for me. The video's director picks the style (whiteboard, chalkboard, notebook, Pixel Quest, Mosaic, or a paper-collage promo when you name a product) and says why: KinoDraw Cloud's GPT-6 Luna, your own AI key, or, offline, word rules on your computer. It only picks a style that works in the format you chose.

Install

App: download KinoDraw for Windows, macOS (Apple silicon, M1 or newer) or Linux from Releases and open it. The first video downloads the doodle search model (about 70 MB) and, when using the built-in voice, the voice model (about 190 MB per language), once, checksum-verified, showing megabytes and percent as it goes.

The app isn't notarized by Apple or signed for Windows yet, so the first launch needs one approval:

• macOS: unzip, drag KinoDraw to Applications and open it. When macOS blocks it, go to System Settings → Privacy & Security and click Open Anyway. If macOS says the app is damaged, run xattr -dr com.apple.quarantine "/Applications/KinoDraw.app" once. Step by step, with pictures: first time opening on a Mac.
• Windows: unzip and run KinoDraw.exe in the KinoDraw folder. If SmartScreen appears, click More info → Run anyway.
• Linux: unpack and run KinoDraw/KinoDraw.

Coming from Doodle Studio? Close Doodle Studio, then open KinoDraw: it moves your Doodle Studio projects, settings and downloaded voice over the first time it opens (if Doodle Studio was still open, KinoDraw says so and finishes the move the next time it opens). Re-enter any saved API keys once (KinoDraw Cloud needs no sign-in), then delete Doodle Studio.app (or the Doodle Studio folder on Windows and Linux).

Command line (Python 3.10–3.12):

uv tool install git+https://github.com/edwardaiwang-svg/kinodraw         # or: pipx install git+https://…
kinodraw studio                                                         # opens the app window
kinodraw make my-script.md -o "My Video"                                # or straight to an MP4

Write a script

Plain text, Markdown or a .docx file works.

• # gives the title; each ## heading becomes a section with its own agenda card. With no headings, the text is split for you.
• Write the way you would speak it. Numbers are read out properly ("$3.8bn", "1450s" and "30%" all work, and so do "2026年" and "百分之三十").
• A final "Conclusion" or "总结" section becomes the ending.

Directors: who plans the visuals

Director	Cost	Notes
Offline (always available)	free	Matches the words of each sentence to the doodle library on your computer. The visual plan stays on your machine.
KinoDraw Cloud	free	GPT-6 Luna plans each section. No account and no key needed; signing in with an email code is optional. Paid plans with more videos and Claude Opus 5.5 come later.
Your OpenAI key (Advanced)	about $0.02 per 15-min video	Default gpt-6-luna. In the app, turn on Settings → Advanced directors to see these last four options.
Your Anthropic key	about $1 per 15-min video	claude-opus-5, claude-opus-5-5, or the cheaper claude-haiku-4-5.
OpenAI-compatible	varies	OpenRouter, DeepInfra, Groq, or a local Ollama or LM Studio.
Your own command	varies	Any program you choose: it gets each request as JSON and prints the plan as JSON.

AI answers are always checked by code before use:

• doodles must come from the library;
• drawing triggers must be words that are actually spoken;
• numbers, dates and quotes must appear in your script.

Any beat that fails keeps the offline plan. Keys are stored in your system keychain.

Your own voice server (optional)

The built-in voice is the default. To use an OpenAI-compatible voice server you already run, open Studio → Settings → Voice server, enter its address and model, optionally its voice and API key, and press Test. Turn on Read scripts with my own OpenAI-compatible voice server and press Save. When it is on, New video and each project's Narrator tab name the server voice, you can choose a server voice per video, and Hear it plays a sample from your server. The voice fields offer the voices your server lists (or, for OpenAI's own API, its documented voices), and you can still type any name. Your own recordings stay local and take priority over the server.

For the command line:

export TTS_BACKEND=openai_compatible
export TTS_API_BASE=http://localhost:8080/v1
export TTS_MODEL=qwen-tts
# Optional: export TTS_VOICE=your-voice
# Optional: export TTS_API_KEY=your-key
kinodraw make my-script.md -o "My Video"

You can also pass --voice-server URL --server-model NAME --server-voice NAME to make, new or voice, and store its key with kinodraw key set voice-server --url URL. A key is kept for the address it was saved for and sent only there; only your own settings choose the server, never a project file someone sends you. To return an existing project to the built-in voice, run kinodraw voice "My Video" --voice-server none and unset the TTS_* variables for future projects.

How it works

script ─▶ ingest ─▶ beats + spoken text ─▶ director ─▶ storyboard.json ─▶ Kokoro voice ─▶ timeline ─▶ renderer ─▶ mix + package
          (md/docx)  ("$3.8bn" → "three point    (rules or LLM)  (editable)     (per-phoneme    (captions,   (hand-drawn     (music, chapters,
                     eight billion dollars")                                    timing)          music)       frames)         QA, thumbnail)

• Voice: Kokoro through kokoro-onnx, on the CPU. Its per-phoneme timing lines drawings and captions up with the words (typically within 0.2 s).
• Drawing: SVG outlines are traced in drawing order by a photographed hand, then the colour pops in. Text is written stroke by stroke from the font's glyph skeletons.
• Quality checks: every video is fully decoded and its frame count checked. Its audio is compared with the mix, and its chapters are verified.

See docs/architecture.md and the storyboard format.

Limits

• The offline director draws a word only in the sense its section is about, and leaves the word undrawn when it can't tell (a wine-making "press" is never drawn as a newspaper). It can't read between the lines the way an AI director can; swap any doodle in the editor, or use an AI director.
• Rendering takes about as long as the video (a 15-minute video takes about 10–15 minutes on a recent laptop). --workers uses more cores.
• Only English and Chinese are supported so far. Full manual testing was done on macOS; Windows and Linux are covered by CI.

Feedback

Press Feedback or a problem? Tell us in the Studio; it also asks in a small card you can close after each video you make. No account is needed, whichever director you use, and nothing is sent until you press Send. Then exactly this goes to KinoDraw Cloud (api.doodlecloud.org): what you typed, your rating, what the video was for and your video link (if you gave them), the two boxes ("KinoDraw may quote what I wrote" and "I am 13 or older"), your email if you typed one, and the app version, your computer type (Mac, Windows or Linux), the language and this installation's install ID. Your email is kept only if you are 13 or older and typed it. What you wrote is quoted only if you ticked the box, and never with your name. To delete it, email [email protected] with the ID that kinodraw cloud-id prints. Prefer GitHub? Open an issue.

Privacy

Offline mode sends nothing anywhere unless you choose your own voice server or send feedback. With KinoDraw Cloud or your own key, only the text of one section at a time (plus doodle names) goes to the AI provider; with Style > Choose for me, one more request per new video sends the title, the section headings and the first 600 characters of the script (with its language and format) so the AI can pick the style. Offline, Choose for me picks on your computer. KinoDraw Cloud does not store script text. With Settings > Voice server on (or TTS_BACKEND=openai_compatible), the text of each part of the script goes to the server you entered, and nothing else does. The request includes the model and voice names and an API key if entered; Test and Hear it send one sample sentence. To offer the server's voices as choices, KinoDraw also asks it for its list of voice names when you press Test and, while the setting is on, each time Settings, New video or a Narrator tab opens, sending only the model name (and the API key for that address, if any). Your own recordings stay on this computer. See the privacy policy.

Licences

• Code: MIT (LICENSE).
• Original doodles, narrator and drawing hand: CC BY 4.0 (LICENSES/ASSETS.md).
• Everything else (fonts, emoji, music, voice model, libraries) keeps its own licence; see THIRD_PARTY_NOTICES.md.
• Your videos are yours. No attribution is required, though "Made with KinoDraw" is appreciated. Videos end with a 2-second "Made with KinoDraw" credit that you can switch off (untick "Made with KinoDraw" end card next to Make video, or kinodraw make --no-credit).