If you have an idea for an explainer or product demo but no animation skills, you can turn a written description into a narrated video. Text-to-animation uses text to create moving visuals. Some tools create short video clips from prompts. explainroo uses an AI coding agent to build illustrated scenes that match the narration.
For narrated explainers, the recommended tool is explainroo, a free, open-source kit that creates an MP4 on your computer. You describe the subject, and the agent writes the narration and draws the scenes. This approach lets you make an informative animated video. explainroo does not generate live-action footage or edit existing camera video.
Two ways a description becomes animation
Generative video tools turn a prompt into a sequence of frames. Many use diffusion, a method that starts with visual noise and repeatedly refines it toward a result that fits the prompt. Video models commonly process the video in a compressed mathematical representation before producing visible frames. This can create striking short shots, but models still struggle to keep scenes continuous, show precise actions, and represent physical details. In a benchmark covering 12 categories of physical laws, tested models scored below 0.60 on average in every category, according to T2VPhysBench.
A narrated explainer uses separate illustrated scenes. The creator breaks the subject into ideas, writes the script, and matches each drawing to the narration. Each scene supports a specific explanation, making the format useful for teaching a process or demonstrating a product.
Make a narrated video with explainroo
explainroo works with Claude Code, Codex, Pi, OpenCode, Gemini CLI, and other coding agents that can run shell commands. It works best with Claude Code. Give the agent this prompt, replacing the bracketed text with your topic:
Make me a short explainer video about [your topic]. Use explainroo for it: clone explainroo, read its AGENTS.md and follow the steps.
The agent sets up explainroo, creates the video, checks it, and provides an MP4 file. It creates two main files: script.md, which contains the narration, and scenes.js, which contains JavaScript instructions for drawing the scenes. A drawing can appear when a chosen word is spoken, syncing the visuals with the narration without a separate manually timed track.
Kokoro, an open voice model, reads the script in a choice of natural American or British English voices. Whisper records when each word is spoken. Chrome draws the frames on an HTML canvas, a digital surface for drawing graphics. explainroo can use hand-drawn-style lines and icons, and scenes can include charts, code, or your screenshots. It adds background music and small sound effects, lowers the music while the voice speaks, and uses FFmpeg to assemble the MP4.
The agent checks scene stills and a contact sheet of frames. It also checks for cut-off or overlapping text and for misread words. Since an agent cannot watch the finished video, review the MP4 yourself before sharing it.
For product demos, tell the agent which product to show and where its code or website is. It can rebuild screens with matching colors, fonts, and button labels, then animate a mouse pointer clicking and typing as the narration explains the product.
You can change the video’s look and format in video.json. The available looks are paper, clean, chalk, blueprint, and midnight. Formats include widescreen, tall, portrait, and square; tall, portrait, and square videos get word-by-word captions. The pace setting adjusts the voice, pauses, and animation speed.
Shape the description before generating
A useful description names the audience, the point to explain, and the visuals that would make that point clearer. For example, for a video about how a plant grows, ask for scenes showing a seed, roots forming, and a sprout reaching toward light. Keep each scene focused on one idea, and check that the narration says what the drawings are meant to show.
Generative video is better suited to short visual moments and experimentation than to tightly controlled, long stories. Benchmarks measure qualities such as motion smoothness, flicker, and whether objects keep their identity over time. As VBench++ illustrates, one impressive frame does not show that the whole clip is consistent. For explainroo, inspect scene timing, legibility, and narration before publishing.
Know what explainroo does not do
explainroo does not edit existing camera footage, make talking avatars, or create AI-generated live-action video. It has no drag-and-drop editor; you ask the coding agent to make changes. Most parts of the kit cost nothing per video, but the coding agent is a separate service with its own pricing. Optional AI illustrations through OpenRouter cost per image.
To run explainroo yourself, install a recent version of Node.js, FFmpeg, and Chrome or Chromium. The first setup downloads the voice and timing models. No graphics card is needed. The project was developed and tested on Linux; macOS and Windows have had less testing. Setup commands and documentation are available in the explainroo docs.
