Claude-real-video － any LLM can watch a video

A new tool called claude-real-video has crashed into the "AI can totally watch video" fantasy with a big reality check: apparently, most chatbots don’t really watch a clip at all. They mostly read subtitles or grab a few snapshots and hope for the best. This project promises to do the nosy detective work on your own computer instead, pulling the important visual moments, skipping duplicate shots, and adding a written version of the audio so you can feed the whole package into Claude, ChatGPT, or Gemini.

But the real show is in the comments, where the community instantly turned this into a mini soap opera. One camp was basically "finally, someone fixed this obvious problem", with the creator openly venting frustration that existing tools miss fast edits and drown in useless still frames. Another crowd jumped in with "cool, but AI still doesn’t really understand motion", saying that for animation and timing, a plain written description may still beat a pile of images. That’s the kind of nerd fight that starts polite and ends with everyone defending their preferred workflow like it’s a family heirloom.

Then came the practical chaos: one commenter imagined using it to track phone charging speed with a camera pointed at a battery meter and temperature gun, which is exactly the kind of gloriously overcommitted experiment the internet loves. Another asked the lurking nightmare question: what about fast scrolling? And one of the spiciest takes wasn’t about the tool working at all, but about the name, with someone arguing it’s too tied to chatbots and could be bigger than that. In other words: useful launch, immediate identity crisis, classic comment-section energy.