Analysing images using AI

Transcript

0:00 Hey everyone, here's a really quick tutorial on using the GPT vision model. So I basically asking questions to or about images tutorial for Greg. 0:10 Uh, Greg, um, do it in English. No, you know, you speak Dutch, but do it in English so other people can. 0:16 Uh, potentially benefit from it as well. So the reason why I'm recording this video, because a, some people may not know that this is a thing and be Greg mentioned, he ran into rate limit error. 0:27 So basically, um, he had like 3000 rows he wanted to run and it was telling him that, uh, he was asking, like he was saying too many requests, not an issue that we found. 0:38 So basically, um, you use the chat, GPT analyze image integration, you use your own API key, and then you have your prompt right here, but you say, okay, you know, use this, uh, image. 0:52 And then, you know, tell me this or that about it, then webpages, your screenshots. So you add those in here. 1:00 Now we, um, have already, like we already have images here. So these are LinkedIn ad creatives. So we have that. 1:10 And then the click URL is, it should look also screenshot that, um, and then these options, leave them toggled off. 1:17 Um, then, uh, I tried sending these as raw image URLs, but they're not. So for it to be a raw image URL, um, can, like, I haven't tried it a lot, but you can pretend to. 1:33 You also take your own screenshots first using, like, a third-party API and then send those. Um, but in my case, just sending the, like, the, the link to where that image was hosted and having ChatGPT take the screenshot that worked, um, and everything else is just pretty much standard. 1:50 Um, so, like, it's small conditions. There's no formula in there, um, about, um, you know, only run it if, you know, there's an actual description, that type of thing, um, and that's everything, so you run that No rate limit issues here, um, so just using the basic setup, then everything gets throttled 2:16 just fine by Clay, um, and as you can see, you know, this runs and it just grabs the images, um, and provides feedback. 2:25 based on the images, so it took this image right here, for example, had a look at it, and then told me what it thought about it, and that's how you can use the, um, uh, analyze image models within Clay.