Warning, this is about AI, so is obviously wrong and awful and no one should read it. Also no one should use it cause AI bad. Also I should obviously be punished cause I use AI so I a bad boy. /s

Google released a new Gemini model, 2.5 Pro. This model, on Google's AI Studio page, can describe Youtube videos. You'll need to use a prompt like:

Create audio description for this video, with timestamps.

The prompt is important because that's probably something like what it was trained on. This is possibly a big hint as to what AI features for accessibility Google is cooking up. Timestamps probably keep it on track more, so they're important. You can also ask it to give narration from the video too if you want to just read the transcript instead of trying to follow along in a more narrative-driven video. And yes, videos are much more visual even in narrative ones, than you'd think. Sometimes, it'll spit out JSON. Sometimes, it'll go off the rails and be completely wrong. And sometimes, it'll be right, and you'll be amazed. Sometimes, it'll refuse to describe a video and you'll be annoyed.

Anyway have fun, and also it's free! So lets boil some oceans for some a eleven y! I mean, unless sighted people *want* to describe every single Youtube video in existence for us, in high quality cause we can't have TTS audio descriptions, oh no. Wait what? People don't want to describe 8 hours of Legend of Zelda gameplay? Or old Dark Shadows episodes? Whaaaat? Welp, AI it is. :)

#AI #google #accessibility #a11y #blind