WEBVTT 00:00.720 --> 00:02.560 Let's do animated frames. 00:03.320 --> 00:03.640 All right. 00:03.640 --> 00:07.920 Let's do spatial prompting or annotated frames in video three. 00:07.960 --> 00:11.040 And this is something I haven't really seen in the other video models. 00:11.080 --> 00:12.440 It's pretty special. 00:12.720 --> 00:18.400 I think just because video three is a really good model, it just understands what's actually in the 00:18.400 --> 00:19.760 frames of the image. 00:19.760 --> 00:24.160 And therefore you can use the frames of the image as a prompting surface area. 00:24.200 --> 00:28.120 So if I just upload this image here, it's one that I made before. 00:28.360 --> 00:33.480 It literally just went online for a simple let me just show you this. 00:34.600 --> 00:44.080 A simple car image and then I'll show you my chairs. 00:45.280 --> 00:45.480 Yeah. 00:45.520 --> 00:49.640 So it's a simple image and I've just a road, right. 00:49.680 --> 00:51.760 And I want a car to drive down here. 00:51.760 --> 00:54.800 I want birds to fly in all directions as the car passes. 00:54.840 --> 00:57.040 And I want an airplane to fly it that way. 00:57.280 --> 00:59.560 And I can put all that in the image. 00:59.560 --> 01:04.320 And Google will understand that as long as you put that as the first frame. 01:04.320 --> 01:07.040 And that's pretty cool that it knows how to do that. 01:07.040 --> 01:12.640 The reason why this is useful, by the way, is that quite often when you're doing video stuff, it's 01:12.640 --> 01:18.320 hard to explain, especially for complex scenes like where you want things to go or like where they're 01:18.320 --> 01:20.640 supposed to enter from, where they're going to be. 01:20.720 --> 01:26.760 Because if I said, for example, I want that tree to have birds fly out in all directions, it might 01:26.760 --> 01:30.000 think it mean one of these trees or this tree on the left. 01:30.200 --> 01:35.080 And I can highlight there and then point out where I want that to happen, which is really cool. 01:35.120 --> 01:35.360 All right. 01:35.400 --> 01:40.520 So let's say a car drives fast on the road. 01:41.440 --> 01:44.480 And we're just going to generate that note. 01:44.480 --> 01:45.960 We haven't said anything about birds. 01:45.960 --> 01:47.440 We haven't said anything about airplanes. 01:47.440 --> 01:50.160 So let's see if they show up in the final video. 01:51.040 --> 01:51.560 Okay. 01:51.600 --> 01:54.080 So we can see our image here and the videos come back. 01:54.080 --> 01:58.880 So let's see if they followed our instructions that remember were not in the prompt. 01:58.920 --> 02:01.000 We just said car drives fast along the road. 02:01.800 --> 02:02.320 Okay. 02:03.120 --> 02:08.320 Pretty much like almost the airplane was flying the wrong direction and the birds fly out there. 02:09.560 --> 02:14.200 The drawing stayed on for a little bit longer than we wanted it to, but otherwise pretty good. 02:14.200 --> 02:16.240 And it's only going to get better from here.