Claude Makes a New Kind of Video - Full Tutorial

Summary

The video demonstrates a way to create motion-graphics and explainer videos with Claude without using a text-to-video generator or installing frameworks such as Remotion or Hyperframes. Instead, Claude writes plain JavaScript that draws each video frame, including visual elements and sound effects. Because the output is code, the creator says it is relatively easy to request precise edits, such as changing colors or adjusting movement.

Style Atlas and Prompt Workflow

• The presenter shares a free database of 86 video styles, ranging from glassmorphism and flat-vector looks to pixel art, chalkboard visuals, and cut-paper dioramas.
• Each style includes a sample and a detailed prompt containing the code used to create it. The presenter argues that this code gives Claude a more reliable starting point than asking it to infer a style from video screenshots.
• To create a video without narration, choose a style, enter a topic or story, and copy the prompt into Claude. You can optionally review a visual storyboard before generating the full video.
• The resulting video can be revised with follow-up instructions. A customized style can also be saved as a Claude skill for reuse.

Voiceover Tutorial

The presenter also demonstrates how to synchronize visuals with narration:

1. Generate narration. For the demonstration, the presenter uses Gemini text-to-speech in Google AI Studio, though ElevenLabs or another voiceover tool can also be used.
2. Obtain a transcript with timestamps. Word-level timestamps offer closer synchronization, while line-level SRT timing may be more practical for longer videos.
3. Choose a style, enable the voiceover option, provide the topic and transcript, and send the prompt to Claude.
4. Upload or drag in the audio file, either with the initial request or afterward, and let Claude match the visuals to the narration.

The example video about how stars are born is described as well synchronized and polished.

Review and Model Comparisons

• The presenter says both Claude Opus and Sonnet produced good results, with Sonnet performing similarly to Opus on the tested prompt.
• ChatGPT also produced a good result, though the presenter judged it slightly weaker than Claude’s.
• The presenter highlights the range of styles and the ability to customize them, while noting that finding and refining a style suited to a channel’s visual identity may take time.

Main Speakers and Sources

• Main speaker: Futurepedia presenter, demonstrating the workflow and sharing the Style Atlas.
• Referenced creator: Kevin, whose posts inspired the presenter’s renewed interest in code-generated motion graphics.
• Tools discussed: Claude (Opus and Sonnet), Gemini text-to-speech, ElevenLabs, and ChatGPT.