Explain Ideas with Cartoon Characters and Scenes
Turn a topic into a one-minute landscape explainer with flat cartoon scenes, animated characters and narration. Finn’s prompt takes Claude from script and shot approval through image generation, voice timing and Canvas rendering.
Requirements
- Python
- Playwright
- Chromium
- FFmpeg
- faster-whisper
- OpenCV
- Pillow
- librosa
Overview
What you get
- The complete author prompt and a Chinese translation.
- A staged guide to scripting, previewing and rendering a narrated cartoon explainer.
Tutorial
Watch the characters act out the explanation
A voiceover alone can leave viewers struggling to picture a process. This prompt asks Claude to turn a topic into illustrated scenes, using character movement, gestures and changing objects to explain it in a roughly one-minute landscape cartoon.
The original video explains Claude. Watch the researchers carrying boxes past an OpenAI building, the robot at a chalkboard and the character sitting at a computer: locations establish context, actions show what happens and signs or screens carry the text. The hosted video is 1280×720 at 60 fps, runs for 61.9 seconds and has audio. The prompt requests a 1920×1080 export with artwork prepared at 3840×2160.
Choose between related explainers
- Turn a Topic into a Paper-Style Cartoon Explainer also starts with a topic, but uses portrait framing, a paper background, a presenter in the corner and word-synced captions. This prompt uses landscape scenes with characters acting out the explanation and text placed on signs or screens.
- Turn Narration into Animation in Seven Styles starts with an existing Chinese recording and offers several visual styles. This prompt starts with a script and voice choice, then builds scenes in one flat cartoon style.
Supply a topic and approve the script and shots
Use a Claude coding session that can read and write local files and run commands. You need access to an image model that can use style references, plus ElevenLabs or Kokoro for narration. Python, Playwright with Chromium, FFmpeg and the other listed tools process assets and export the video. Claude checks these tools and installs missing dependencies as instructed in the prompt; there is no separate skill package to install.
- Create a video project directory, open Claude there and send it the complete prompt from this page.
- Answer its questions about your topic, image model and voice service. Use “Why sync is not backup” for this exercise. The practice requests below were written by the Opuskill editors.
Topic: Why sync is not backup
Audience: People who sync work files across devices using cloud storage
Takeaway: Deletions can spread to other devices; an independent backup can preserve recoverable historical copies
Facts to keep: Sync makes files accessible across devices; syncing a deletion does not preserve history; recovery depends on the actual backup settings
Use English narration, targeting 55–65 seconds
Explain with an office, two computers and an independent backup cabinet; do not invent product features or statistics
Image model: [the model you can access]
Voice service: [ElevenLabs or Kokoro]
Show me the script and shot plan first, explaining what each scene demonstrates; wait for approval before generating assets
- Check that the script explains the difference between sync and backup, and that the shots show the relevant file changes. For example, show the same file on two computers, show a deletion spreading, then restore an independent historical copy.
- Approve the facts, shot order and pronunciation before generating assets. The prompt’s 190 words per minute is an English pacing reference; use the actual recording to establish the duration in other languages.
The original structure includes origins, dates and categories. If those do not help explain your topic, ask Claude to adapt the sequence around the necessary steps rather than adding history or numbers to fill a slot.
Prepare scenes, characters and narration
The image model draws the backgrounds and characters. Code animates them, writes on signs and schedules actions against the narration. Generating backgrounds and characters separately lets you change a gesture without repainting the scene.
- Start with an empty office background for one shot, leaving computer screens and signs blank. Generate characters on solid green. For a character who gestures often, leave shoulder sockets so code can draw the arms.
- Inspect the assets: check for unwanted people in the background, poses that do not fit the shot and green clothing that would disappear during keying. The prompt removes green regions connected to the image border to avoid cutting into dark clothing.
- Generate the opening sentence as a voice preview. Approve the language, pronunciation and voice before generating the full narration. If using a voice reference, supply a narrator sample you have permission to use.
- Listen to the final audio, then ask Claude to obtain word timestamps with faster-whisper. The prompt also calls for shortening long pauses and adjusting the overall speed. Approve the processed recording before using it to time the shots.
When narration changes, regenerate the affected audio and update word timings. Cuts, text and gestures must follow the final recording.
Make one scene clear before expanding
Finish the “deletions sync” segment first so you can judge visuals, movement, text and sound together. Send this request in the same Claude session:
Complete only the “deletions sync” segment using the approved final narration
Keep the landscape composition and flat cartoon style without outlines
Place two computers in an office, with file names written on their screens
A character deletes a file on the first computer; the second computer shows the deletion when the narration mentions syncing
Show key frames and a preview with sound; wait for approval before building the other scenes
- Inspect the key frames: is text on the screens and signs within the scene, are files visible around the characters and are there any black edges?
- Play the preview: does the deletion match the spoken explanation, do gestures reinforce the same meaning and is the sequence easy to follow?
- Give feedback with an object and a timing cue, such as “The second computer deletes the file too early; move that action to the word ‘sync’.” Approve this segment before expanding the film.
The author specifies one HTML Canvas whose frames are determined by time, with window.seek(t) for seeking to a moment. Playwright drives browser frame capture and FFmpeg encodes the result. Heads, arms and legs move independently; backgrounds and characters share the same camera transform to keep them aligned.
Inspect the whole film and export
- Once all shots are assembled, ask for the two 16-frame contact sheets specified in the prompt. Check for blank boards, overflowing text, covered faces and black edges.
- Correct the affected shots, then play the whole film with sound. Listen for clear narration and check that cuts follow changes in meaning rather than a rigid interval.
- Approve the render. The prompt specifies 60 fps, motion-blur sampling, H.264 at CRF 16 and a final mix of -14 LUFS. Ask Claude for the actual render command used by the project and the exported file path.
- Open the export and check its duration, aspect ratio, ending and audio. Follow the licence of the music and effects you actually select.
The post names Pixabay for music and Mixkit for effects. Choose the assets first, then align typing, wipes and red-cross effects with their visual events while keeping the voice clear above the music.
Change one scene, then use your own topic
Start with a visual change to understand how it differs from a narration change:
Replace the independent backup cabinet with a wall-mounted archive shelf in the office
Have the character take out a folder with an earlier date to demonstrate restoring an older version
Keep the approved script, audio and shot duration
Show the shot’s key frames before and after the change, then provide a preview with sound
A scene-only change can keep the voice timing. Replacing a spoken sentence requires new audio, updated word timings and a check of the shot length.
For your own topic, replace the brackets below and send the request together with the complete prompt:
Topic: [one question to explain]
Audience: [what they already know and what they misunderstand]
Takeaway: [one concrete conclusion]
Evidence: [sources, links or file paths]
Available image model and voice: [the services and language you can actually use]
Scenes: [places, people and objects that explain the process]
Duration: [a target within 55–65 seconds]
Approve the script and shots first, then assets and voice, then one scene, then the whole film
Each scene should answer “What does the viewer understand here?” Arrange actions and text around that answer. You can then reuse the production method without copying the original video’s locations and characters.