WEBVTT 00:00.050 --> 00:06.470 In this video, we're going to take a look at a website that can help detect deepfakes and altered images. 00:07.180 --> 00:09.250 Now the website is sensitive. 00:09.700 --> 00:11.730 It can be found at platform sensitive. 00:12.130 --> 00:12.530 I. 00:12.640 --> 00:15.280 Forward slash deepfake detection. 00:16.190 --> 00:22.880 Now it's primarily designed to find deepfakes, which you could either upload a video, you could or 00:22.880 --> 00:23.880 you could take a still photo. 00:23.900 --> 00:27.860 This also I found works with things like AI generated photos. 00:27.860 --> 00:30.860 It kind of uses, I guess, the same technology behind it. 00:31.880 --> 00:33.380 And this could help detect it. 00:33.380 --> 00:36.230 And when I say help, it means that, well. 00:37.080 --> 00:39.630 It might not be 100% accurate. 00:39.660 --> 00:45.120 The samples I ran through here were did actually detect it at a pretty good rate. 00:45.930 --> 00:48.690 And this, of course, could help you with things like. 00:49.220 --> 00:55.220 Well, if you're doing an investigation, if you're doing something like taking a look at potentially 00:55.220 --> 01:02.570 a fraudulent profile, say a dating profile, something else, something that might be a little suspect, 01:02.570 --> 01:09.020 well, you could run the photo or the URL or the video through here and potentially find if it's was 01:09.020 --> 01:13.700 created with a Deepfake or other AI generated type program. 01:14.060 --> 01:17.060 Now, this isn't really designed for things like Photoshop. 01:17.270 --> 01:19.850 Again, this is going to be a deepfake detector. 01:20.830 --> 01:26.260 And if you're coming from the ocean one or Ocean two, course you may find a duplicate in here. 01:26.260 --> 01:31.660 I thought it was a really cool and important program, so I'm actually adding in both courses. 01:31.870 --> 01:33.460 So let's take a look. 01:34.540 --> 01:40.780 So you are going to need a Google account or to create an account when you run this program and it is 01:40.780 --> 01:41.080 free. 01:41.080 --> 01:42.490 So don't worry about that. 01:43.340 --> 01:49.190 So I have the I have the website open here and we can take a look at different things. 01:49.220 --> 01:53.930 Again, you can either drag or drop or browse files and add a file in there. 01:53.960 --> 01:58.730 Again, it could be a photo, it could be a video or a URL, and the URL has to be on their list in 01:58.730 --> 02:00.770 order for it to be able to run it. 02:00.770 --> 02:04.580 So if you have a URL, it may or may not find it. 02:05.910 --> 02:06.360 Again. 02:06.360 --> 02:08.040 That's why it's always best to just. 02:08.740 --> 02:09.580 Upload the file. 02:09.580 --> 02:20.290 If it doesn't detect it and we can see the available formats png jpeg jfif tiff mp4 and mov file. 02:20.290 --> 02:27.130 So it has a pretty wide variety and the video limits are going to be 30 megs and up to ten minutes at 02:27.850 --> 02:33.940 1440p So again, pretty high resolution, pretty good video length. 02:34.030 --> 02:36.340 Ten minutes is pretty good. 02:37.090 --> 02:39.160 And again, we could use this for things like. 02:39.940 --> 02:42.450 Well, is this misinformation? 02:42.460 --> 02:44.160 Is this a fake dating profile? 02:44.170 --> 02:46.780 Is this a fake profile trying to scam someone? 02:47.020 --> 02:47.560 ET cetera. 02:47.560 --> 02:48.190 ET cetera. 02:48.190 --> 02:49.000 So. 02:50.790 --> 02:56.700 I took a couple photos from this person doesn't does not exist and I ran it through the program. 02:56.700 --> 02:59.850 Now, depending how long it takes, is going to vary. 02:59.880 --> 03:04.830 It actually took a few minutes when I ran this, so that's why I ran a few ahead of time. 03:04.830 --> 03:07.860 So let's take a look at the at the detection rates here. 03:08.370 --> 03:13.460 So in here we can see it tried to detect a face swap and it was unable to detect it. 03:13.470 --> 03:15.450 However, on the. 03:16.040 --> 03:24.560 The Gan generated face, it was able to detect a that this was fraudulent and it had 99.9% confidence 03:24.560 --> 03:24.920 in that. 03:24.920 --> 03:25.670 So it's nice. 03:25.670 --> 03:31.220 It gives you a an actual rating for how confident it is that this is a fake photo. 03:33.100 --> 03:36.700 And we could see the type of model generated that was used. 03:36.730 --> 03:41.560 Again, here's another one, 99.9, and this one was 89.2. 03:41.560 --> 03:46.270 Again, not as high as other ones, but still pretty high detection rate. 03:47.190 --> 03:52.800 Now, it's not going to be able to do things like if a person has a mask on it, it does have a hard 03:52.800 --> 03:56.970 time being able to determine whether that's a legitimate photo or not. 03:58.440 --> 04:03.600 And other things we could do are we can click on intelligence here and we can we can go through here 04:03.600 --> 04:06.780 and we can see various things on here. 04:06.780 --> 04:09.060 We can see the watch list was popular. 04:09.060 --> 04:11.460 That's being generated around. 04:11.730 --> 04:18.900 And we can we can actually type in public figures and see if that was actually if it's coming up on 04:18.900 --> 04:19.350 the site. 04:19.380 --> 04:25.260 Now, the most targeted in entertainment last month, we could see different, different people in here 04:25.260 --> 04:28.350 like Gal Gadot, and we could see who they are. 04:28.380 --> 04:33.090 Vulnerabilities Found Month How much has grown in the past month and risk rating. 04:33.960 --> 04:34.980 So. 04:35.950 --> 04:40.360 Obviously, if you're looking for a celebrity, this is a pretty good place to go to see if there is 04:40.360 --> 04:43.930 a what type of detections there are here. 04:44.110 --> 04:49.840 And we can actually if we click on any of these, if you're investigating a public figure, we could 04:49.840 --> 04:54.760 potentially use this click on here, go through here and we can see the different detections. 04:54.760 --> 04:59.110 And again, we can see Deepfake on here. 04:59.110 --> 05:04.600 And if there's a particular date, we can try to go to the source and determine that whatever we're 05:04.600 --> 05:10.380 searching for, if it was actually a deepfake or not, or again, we always can go back in there, upload 05:10.390 --> 05:18.280 a file or video or the URL, run it and see what the callback is, whether it's high confidence rate 05:18.280 --> 05:21.070 or low confidence rate or if it was a face swap. 05:21.680 --> 05:23.780 So again, pretty cool website. 05:23.780 --> 05:26.170 It's free to use sensity. 05:26.210 --> 05:28.130 It's platform sensitive. 05:28.550 --> 05:36.170 I dot forward slash deepfake detection and I'll put a link into the this website in the description 05:36.170 --> 05:36.980 of this video. 05:37.010 --> 05:38.090 Thank you for watching. 05:38.090 --> 05:38.870 I'll see you next video.