Opus 5.5 made my entire last video. The script, the 3D graphics, and the edit. I recorded 42 minutes of retakes, stutters and me saying "what the hell" at my own slides. I gave it one prompt per recording. It came back as an 11-minute video, with 14 retakes cut down to the final take. The part that surprised me most: I told it that it zoomed in too much. It told me it didn't. My Mac's accessibility zoom was on while I recorded. Then it rebuilt that whole minute from my slides. I also put it side by side with GPT-6 Sol on the same job. It's not even close. Here's the cut prompt so you can try it today: "Cut it tight, especially the intro. Cut all the retakes out (first 1 or 2 takes, leave the final take in), filler words, stutters, ums, ahs. Add sound effects to topic transitions. Zoom in on mouse hovers." Watch the full breakdown here: [https://youtu.be/_akY3ktZK6Q] Two asks: 1. If it helps, leave a comment on YouTube in the first few hours. That's what gets it shown to people outside this group. 2. Try the prompt on one recording you've been putting off, and post your before → after length in this thread.

🚨 New video is live! This one's a bit different from my usual builds. Less "new model dropped" and more the stuff under the hood that nobody really explains. I followed one prompt from the second you hit Enter → into a data center → and back to your screen. All in 3D. 5 words everyone throws around: token · parameters · training · next token · context window Each one explains something weird AI does. Like why it couldn't count the r's in "strawberry", or why your long chat slowly gets worse. And each one changes how you prompt. The last part is the practical one: the 1–2 sentences to add so it stops guessing what you want. [VIDEO LINK] watch + drop a comment on YouTube, show the support there! 💪 Honest answer: which of the 5 words could you actually explain before this?