WEBVTT 00:00.180 --> 00:01.020 -: All right, welcome back. 00:01.020 --> 00:02.400 So in this video, we're gonna have a look 00:02.400 --> 00:04.320 at how you can automate product descriptions 00:04.320 --> 00:06.750 using OpenAI's Vision. 00:06.750 --> 00:08.130 We're gonna have a look at, see, 00:08.130 --> 00:09.480 the kind of package imports 00:09.480 --> 00:12.090 you'll be importing is OpenAI, os, 00:12.090 --> 00:13.440 and then we're gonna use text wrap 00:13.440 --> 00:16.920 to easily see the product descriptions that we get back. 00:16.920 --> 00:19.260 If you don't have an OpenAI API key, 00:19.260 --> 00:20.940 you'll need to insert one here 00:20.940 --> 00:23.100 and then we're gonna load a text_wrapper. 00:23.100 --> 00:25.200 And to start with, we'll use a prompt 00:25.200 --> 00:28.290 to specifically say, Act as a fashion retailer. 00:28.290 --> 00:29.520 You're responsible for writing 00:29.520 --> 00:31.380 effective product descriptions. 00:31.380 --> 00:33.870 We're gonna have a look at this black navy trousers 00:33.870 --> 00:35.760 and see how we can automatically generate 00:35.760 --> 00:37.290 product descriptions for that. 00:37.290 --> 00:40.410 So what we're gonna use is the OpenAI's Vision API, 00:40.410 --> 00:43.770 and you'll see we have to use this gpt-4-vision-preview. 00:43.770 --> 00:45.360 We've also got some interesting things. 00:45.360 --> 00:47.370 So you can see here there's this image_url, 00:47.370 --> 00:49.470 which is the image URL I just showed you. 00:49.470 --> 00:50.490 And we've got our prompt 00:50.490 --> 00:53.580 that's specifically telling the ChatGPT API, 00:53.580 --> 00:55.200 in this case the Vision API, 00:55.200 --> 00:57.000 that we want to get a product description 00:57.000 --> 00:58.740 from this specific image. 00:58.740 --> 01:00.510 Now, after loading the client, 01:00.510 --> 01:02.640 we can then see what kind of product description 01:02.640 --> 01:03.570 we get back. 01:03.570 --> 01:05.100 And you can see it takes around 01:05.100 --> 01:07.800 about maybe four to 10 seconds on average 01:07.800 --> 01:08.633 to use the vision API. 01:08.633 --> 01:09.840 It can take a little bit longer. 01:09.840 --> 01:10.980 As you can see here, 01:10.980 --> 01:13.410 we're clocking in at about 11.6 seconds. 01:13.410 --> 01:15.720 Now, we have got a product description 01:15.720 --> 01:18.480 for this specific pair of navy pants 01:18.480 --> 01:20.490 in the suit section of ASOS. 01:20.490 --> 01:21.570 But in particular, 01:21.570 --> 01:23.460 I'm not particularly happy with this prompt. 01:23.460 --> 01:25.350 And the reason why is it's quite large 01:25.350 --> 01:28.053 in terms of the product description. 01:29.010 --> 01:29.970 So we wanna limit this 01:29.970 --> 01:31.680 to maybe two to four sentences in length. 01:31.680 --> 01:35.310 So we we're gonna do is we're gonna add some examples. 01:35.310 --> 01:37.950 In this case we're not gonna have a product context, 01:37.950 --> 01:39.180 but that could be interesting 01:39.180 --> 01:41.040 if you needed to be more specific. 01:41.040 --> 01:43.170 That could be something you could add into the prompt. 01:43.170 --> 01:44.760 So you'll now see our improved prompt 01:44.760 --> 01:46.680 is acting as a fashion retailer. 01:46.680 --> 01:48.300 We're asking it to use all the images 01:48.300 --> 01:50.220 because ChatGPT vision API 01:50.220 --> 01:51.960 can accept multiple images 01:51.960 --> 01:53.910 and also we've got some examples 01:53.910 --> 01:55.260 which we've specified earlier 01:55.260 --> 01:57.750 that really helped clarify the length and tone. 01:57.750 --> 01:59.460 We've also got a rule section, 01:59.460 --> 02:00.900 so you could add a rule section here 02:00.900 --> 02:02.070 so that the product description 02:02.070 --> 02:04.440 must be between two to four length 02:04.440 --> 02:05.940 in terms of sentences 02:05.940 --> 02:08.460 and also it must be in a professional tone. 02:08.460 --> 02:10.590 We've also said use the following images. 02:10.590 --> 02:11.850 So let's have a look at this. 02:11.850 --> 02:13.410 So this is a single image, 02:13.410 --> 02:15.240 but you'll see that after this comes back 02:15.240 --> 02:17.280 already I've got a cash response here 02:17.280 --> 02:18.570 and here's another one. 02:18.570 --> 02:21.660 The actual product description length is a lot less 02:21.660 --> 02:23.760 and we've still got some good pits 02:23.760 --> 02:26.490 specifically talking here about the cargo trousers. 02:26.490 --> 02:28.440 So this is a different image. 02:28.440 --> 02:30.458 What we are looking at here 02:30.458 --> 02:32.100 is specifically the cargo trousers. 02:32.100 --> 02:34.440 And that can be interesting from our point of view 02:34.440 --> 02:35.610 because what we're trying to do 02:35.610 --> 02:37.860 is get a reduced amount of size. 02:37.860 --> 02:40.560 Now I want to go and take it another step further. 02:40.560 --> 02:42.870 And what I want to do is I want to say 02:42.870 --> 02:44.550 for this product context, 02:44.550 --> 02:47.760 the item is gonna be a cargo trousers in black 02:47.760 --> 02:49.200 and it's not gonna be a hoodie. 02:49.200 --> 02:51.990 And what we are then saying is act as a fashion retailer. 02:51.990 --> 02:53.550 And so same kind of prompt, 02:53.550 --> 02:56.010 we've got our examples in here, we've got our rules. 02:56.010 --> 02:58.650 The product context is now no longer an empty string. 02:58.650 --> 02:59.790 What's interesting about this 02:59.790 --> 03:02.520 is when you look at cargo pants in particular, 03:02.520 --> 03:05.430 you'll see that the cargo pants themselves, 03:05.430 --> 03:09.210 they actually have a top as well, a hoodie top. 03:09.210 --> 03:11.040 And we wanna make sure that ChatGPT 03:11.040 --> 03:13.290 is really not going to talk about the hoodie at all. 03:13.290 --> 03:15.360 And you'll see in this separate model picture, 03:15.360 --> 03:18.210 we really want to talk about these black cargo trousers. 03:18.210 --> 03:20.100 And so that's why you might have a section 03:20.100 --> 03:21.510 like the product context 03:21.510 --> 03:24.780 to say the product description is this cargo trousers, 03:24.780 --> 03:27.840 it's not the hoodie or some type of product title. 03:27.840 --> 03:29.940 And the other interesting thing as well 03:29.940 --> 03:31.710 is notice how we've put both the images 03:31.710 --> 03:34.770 that I just showed you into ChatGPT, vision API. 03:34.770 --> 03:36.860 So you can have multiple images 03:36.860 --> 03:40.200 to really refine those products description outputs. 03:40.200 --> 03:42.510 So let's have a look and see how long this takes. 03:42.510 --> 03:45.360 Probably should take about five to seven seconds. 03:45.360 --> 03:46.830 So again, you've got five seconds 03:46.830 --> 03:48.000 and if we have a look, 03:48.000 --> 03:49.710 it's roughly about the length that we want. 03:49.710 --> 03:52.530 You've got a nice bit here around the leisure wardrobe. 03:52.530 --> 03:54.600 We've got the ultimate blended comforts 03:54.600 --> 03:56.400 and perfect for those on-the-go moments 03:56.400 --> 03:57.270 or relaxing weekends. 03:57.270 --> 03:59.610 So I'm really happy with how this product description 03:59.610 --> 04:00.480 has turned out. 04:00.480 --> 04:02.040 So hopefully this gives you a bit of insight 04:02.040 --> 04:04.680 into how you could use this at e-commerce websites 04:04.680 --> 04:08.490 to automate your product descriptions using ChatGPT, API.