How to create awesome videos with Claude?
A major part of my 1 GB monthly internet plan in the early 2010s was spent watching Khan Academy videos. I loved the intuitiveness and detail with which Sal Khan explained maths and physics concepts. It helped me build a strong foundation.
Khan Academy was replaced by NPTEL and MIT OCW during my engineering years.
I have always found myself to be a visual learner, and video lectures—with the option of pausing and rewatching—have done wonders for my understanding.
Incredibly, AI models are now advanced enough across modalities to create such videos on any topic, at any abstraction level and with the exact style of delivery that I want.
I asked Claude to help me understand the AI training stages better in Khan Academy style, and here is what it gave me.
How can you do this?
- Use Opus 5.5 (or the latest Opus model at the time of reading) for the best results.
- Start with a style guide—define the background colour, the personality of the narrator, and the style of delivery (pen on screen in this case, but you can also try animated characters, moving objects, etc.). It helps to give a reference video so Claude knows what you’re going for.
- Ask Claude to draft a frame-by-frame plan and script. This is how your video is going to look and sound. At this stage, you can give it inputs and ask it to clarify whatever is unclear. A good way to understand complex topics easily is to ask Claude to use ASD-STE100 as a writing style. It uses simplified technical language, which improves comprehension.
- Once the script is finalised, you can use Google AI Studio Playground’s TTS (text-to-speech) service to generate a human-like voiceover for your script. Download the .wav (audio) file and provide it to Claude. This step can be automated via the API if you are technically savvy. Alternatively, if you don’t want the fuss, Claude can generate a voiceover with a built-in text-to-speech tool. You will just have to bear with a conspicuously machine-generated voice for the duration of the video.
-
Instruct Claude to use JavaScript with Remotion to make the video. I asked Claude to summarise what goes on behind the scenes, and here is its response:
Remotion is a library for making videos with React.
In Remotion, a video is a React component, and the component is told one thing: which frame it’s on. Frame 0, frame 1, all the way to frame 1,800 for a minute at 30 frames per second. From that number, the code works out what the screen should look like. At frame 90, this character is 40% faded in. By frame 120, the cup has slid 12 pixels to the left. Between frames 212 and 280, this subtitle is showing. Every movement is a small piece of maths on the frame number, and Remotion has helpers for easing and springy motion.
Claude draws the characters and props as SVG, which means shapes described in code: a circle here, a curved line there, filled with one of the style guide’s colours. Asking for “the same person, now holding a kettle” means Claude reuses the person’s code and adds a kettle. While working, it opens Remotion Studio, a browser-based preview, and scrub through the video like a normal player.
Nothing is filmed, and nothing is edited. To make the final file, Remotion opens the page in a headless browser (Chrome with no window), takes a picture of every frame, and stitches the pictures and the audio into an MP4.
Once the voiceover exists, a speech-to-text model listens to it and returns a timestamp for every word. That turns timing into data: a subtitle is “show these words from 7.1s to 9.3s,” and a scene changes on a particular spoken word. Because it all comes from the real audio, nothing drifts out of sync. A quiet background track and a pass to a standard loudness level finish the sound.
- That’s it—you have your video! You can go ahead and ask Claude to change the look and feel, explain a part in more detail, or design something differently. As long as the script doesn’t change, Claude makes changes quite rapidly. Any changes to the script require the TTS voice to be regenerated.
Khan Academy’s mission is “a free, world-class education for anyone, anywhere.” I didn’t expect the “for anyone, anywhere” part to become this personal.