D
Dynamo.sh open-source whiteboard video
GitHub Open the studio
Open source · self-hosted · no per-video pricing

Explainer videos that draw themselves.

Give it a prompt, a PDF, or your own voice. Dynamo.sh writes the script, composes each board, and animates a hand drawing it — synchronised to the word being spoken. Then it hands you the timeline to edit.

One video on us, then add your own Groq and Fish Audio keys — both free to create — and keep going for what the providers charge you, which is pennies. Or clone it and run the whole thing on your own machine, where there is no limit at all.

Made by this engine from one sentence: “solve 2x + 6 = 20, step by step”.

What you get that the hosted tools don't

The commercial whiteboard-video tools are good. They are also subscriptions that meter your output, keep your source material, and hand back an mp4 you cannot open. This is the other trade.

Typical SaaS

  • Per-video or per-minute credits
  • Watermarks until you upgrade
  • Your PDFs and prompts live on their servers
  • A fixed set of styles
  • Regenerate the whole video to fix one word
  • API is an enterprise add-on, if it exists
  • You get an mp4 — nothing else

Dynamo.sh

  • Unlimited videos — you pay your model provider, nobody else
  • No watermark, ever
  • Runs on your machine; your documents and prompts stay on it
  • 11 styles from two composable axes, plus your own hand and fonts
  • Edit one element and re-render — no second generation
  • The whole product is the API. 46 endpoints, documented
  • You get the mp4 and the timeline JSON that made it

Built like a tool, not a demo

🎬

Ink follows the voice

Every drawing is anchored to a phrase in the narration. The stroke starts a beat before its word so the ink lands as you say it — the detail that separates a whiteboard video from a slideshow.

🎙️

Your voice, or none

Upload a recording and the boards re-time to what you actually said. Add a transcript and the timing becomes exact. Or record on webcam and appear as an inset beside the visuals.

✏️

A real editor

Drag elements on the canvas, drag bars on the timeline, or just ask in chat. Every change is a validated operation applied to a document you can read, diff and version.

🎨

Looks that mean something

Chalkboard, graph paper, kraft, ruled pad, sharpie. Each is a complete system — surface, stroke texture, palette and handwriting — not a colour swap on the same template.

📐

Composed for the shape

Vertical for Shorts, horizontal for YouTube. Columns collapse into rows, flow diagrams run downwards, type resizes to the edge that runs out first. Not a letterboxed crop.

🔌

Your models, your keys

Groq, Anthropic, OpenAI or a local Ollama. Your own image API for photographs. Built-in system voices, or a cloud one. Swap any of them without touching the rest.

🔎

200,000 icons, no key

Name a thing and it is found — from a built-in set drawn stroke by stroke, or searched across open icon libraries that need no account. Only the word itself is sent, never your prompt or documents, and you can switch the lookup off entirely.

📡

Watchable before it is finished

The board for scene one is settled the moment its narration is spoken, so there is no reason to wait for an encode of the whole thing. Scenes stream out as they are made — first one on screen in seconds — and what travels is the timeline rather than the pixels, so your own page draws it at its own resolution. Encode a file later, or never.

Maths written, not listed

Working stacks down the board the way a teacher writes it — each line under the last, earlier lines staying put across scenes so the reader can look back up. The rule being applied sits in the margin, a tick or a cross goes on the line it judges, and the pen draws one thing at a time rather than three at once.

How it works

  1. 1

    Say what to explain

    A sentence, a PDF, a pasted article — or all three.

  2. 2

    The model directs

    It writes the narration and places every element on the canvas: position, size, colour, draw order.

  3. 3

    A browser draws it

    Frame by frame, deterministically, straight into ffmpeg. No frame is ever guessed.

  4. 4

    You edit it

    Chat, drag, or type exact numbers. Re-render costs one render, not one generation.

What it costs

There is no per-video price here, because there is no per-video cost to us beyond a machine. What a video actually costs is the model that writes it and the voice that reads it, and both of those are billed to whoever owns the key. So that is exactly where the line is drawn.

Your first video

Free

On our keys, on our machine. Enough to see whether the thing does what you need before you set anything up.

Create an account

Self-hosted

No limit

Clone it, run ./setup.sh, and it is yours: no accounts, no allowance, no telemetry, and your documents never leave the machine.

Get the source

Make one and see.

An account takes a moment, and the first video is on us.

Start free Compare it with the alternatives

Or run it yourself and skip all of this — it is the same code either way.