AI's 'NanoBanana' Moment: Video Generation Reaches New Heights!

Transcript

1. 0:00The question was um [clears throat] I mentioned the Omni model as being under hyped. The The person that um that announced the new
2. 0:15Gemini Omni model which is this it it outputs video um was Demis Hassabis. And so Demis Hassabis is the creator of DeepMind
3. 0:28which was acquired by Google. He won the Nobel Prize for chemistry um for AlphaFold for the protein folding um
4. 0:36work that they've done at DeepMind. And And so DeepMind is up up to some incredible stuff. Demis Hassabis has been
5. 0:43thinking about AI and AGI and world models since he was a little boy. Like literally since he was a little boy.
6. 0:50This is This is kind of his life's work. Um How people are kind of characterizing the new video output of the Gemini model
7. 1:02is as uh it's kind of like nano banana it's the nano banana moment for video. So if you haven't used nano banana it's the
8. 1:15image generation tool within Gemini and and it does text really well and it does it's just it's just really good, right?
9. 1:22And then the new image model inside ChatGPT is also really good. And And this new video model from Gemini um
10. 1:31does things really well. It does text really well. It does concepts really well. The whole model is both a large
11. 1:39language model. It's It's a true multimodal model. It's the first one that I know of and I could stand corrected, but I don't
12. 1:47think so. This is the first model that's anything in anything out. So you could put in a PDF and say turn this PDF into
13. 1:55a video. You could put in a um video and say, "Make this video underwater." And it will it will do that. The reason that I think it's
14. 2:04under-hyped is that we now have a model that that understands all of what's in the large language model, right? But it
15. 2:13also seems to understand how the world works, how things interact with one another, and how those things relate to
16. 2:21the concepts that sit underneath them. So, it understands physics, and it understands math, and it understands
17. 2:28history, and it understands art and culture. It understands lighting and shadows. It it it understands how the world works.
18. 2:36And so, if you think about all of the things that you can do with a large language model, where if you give it
19. 2:40more context, if you tell it about a company that you're working with, you could now say, "Why don't we make a video of
20. 2:55Let's say some worker has an idea for a new way of manufacturing something. Let's make a video of what that might
21. 3:02look like on the factory floor." And make like an explainer video of that concept. And then you could say, "Okay,
22. 3:11now um now make it a spreadsheet that that projects the the costs and profit if we took that approach."
23. 3:20And you know, here's all the raw here's all the raw data. And it'll it'll create that. Now make a video of, you know, someone talking about
24. 3:30the financials as if it were in a quarterly meeting. Right? You can create videos of architecture concepts that [laughter]
25. 3:43are derived from blueprints. Um and have them be accurate, right? Um So, so one my go-to um examples that I've always done with
26. 3:57image models was a '70s muscle car in an abandoned factory, some version of that. And whenever I've done the videos, the
27. 4:05videos are just not good. The wheels don't roll right, the wheels turn the wrong direction, the car goes backwards
28. 4:11instead of forwards, things like that. So, this [clears throat] was just a very simple prompt about a '69 GTO Judge um
29. 4:18that that revs its motor, it's driven by someone mysterious, and it leaves the factory. And so, this is what it created. So, pretty good.
30. 4:25Um here's some little documentary stuff. Deep in the lush rainforests of the Lesser Antilles lives the incredibly rare pickle bird.
31. 4:33Its remarkable camouflage makes it nearly impossible to find. >> Yeah, his little buddy. >> more elusive is the Skittle's
32. 4:41salamander. Its skin is adorned with a dazzling array of bright spots. They serve as a brilliant and effective
33. 4:47against the colorful backdrop of the jungle. So, just just like it did all the audio, it it invented the creatures,
34. 4:54it did the editing. So, so just seems to understand things quite well. Watch the full replay at community.thesalon.ai.