How to give a video to Claude

If you try to attach a video to Claude, it won’t work: Claude accepts documents and images, not video files. The good news is that Claude is excellent at reading screenshots and transcripts, and that’s all a video really is.

What Claude can read

Claude handles PDFs, text files and images (JPEG, PNG, GIF, WebP). So the goal is to turn your video into images (what you see) and text (what’s said), and to tell Claude how they fit together.

Step by step

1. Pick the moments that matter. For a screen recording, a screenshot every 2–5 seconds works well. For a talk, only take one when the slide or scene changes.
2. Transcribe the audio with timestamps.
3. Write a short index like “Screenshot 3 (0:15): the speaker says…”. This is what lets Claude connect the picture to the words.
4. Attach everything to one message and ask your question.

The fast way

Unreel builds that whole package for you in the browser. Drop in a video (or paste a YouTube link), pick how to split it, and download a folder with:

• frames/: the screenshots, named with their timestamps
• sheets/: grids of 9 screenshots each, so you upload fewer images
• transcript.txt: every word, with timestamps
• README_for_AI.md: the index that matches each screenshot to what’s being said

Attach the README and the images (or the contact sheets) to Claude and start with: “Read README_for_AI.md first.”

Good things to ask

• “Turn this tutorial into a numbered checklist.”
• “What are the three main arguments in this talk, with timestamps?”
• “Look at screenshots 4–9. What changes in my form between them?”