Images, video and voice

What AI can see, hear and generate beyond text.

What it is

Multimodal models take in and produce more than text: they read charts and handwriting, transcribe and speak, and generate images and short video clips.

Why it matters

Photograph a whiteboard and get notes; narrate a demo video; sketch a concept. It's also where deepfakes and consent issues show up first.

Key ideas

• Understanding an image is different from generating one.
• Generated media should be labeled.
• Never generate realistic media of real people without consent.

Try it: From whiteboard to study sheet

1. Photograph a page of handwritten notes.
2. Ask an assistant to transcribe and organize it.
3. Mark every transcription error it made.