Affiliates have to come up with tons of creatives on a regular basis. However, marketers, end-users, and moderation bots perceive ad images and copies differently. There is no such thing as an objective viewpoint.
Each image is the sum of its elements that can both improve and ruin your conversion rate. However, affiliates are not always able to decipher what exactly is displayed on an ad banner. Fortunately, there are a few AI solutions available on the Internet that can help you better understand your images.
AI can detect multiple objects within your creative for you to test out different image elements and their combinations and get representative data on how effective they are. Testing is something all affiliate marketers are familiar with. So, in this article, we are going to get you covered on what AI solutions to use to break down your creatives and derive insights from your ad images.
Google Vision
The first tool in our list is Cloud Vision API brought to you by Google. To get valuable image insights, the only thing you need to do is upload your creatives through the solution’s easy-to-use graphical interface.
Google Vision detects faces, emotions, and separate objects within your image. It can also read printed and handwritten text and see the dominant colors. Moreover, Vision analyzes the general attributes of the image and generates word clouds associated with whatever is displayed in your creatives. In the “SafeSearch” tab, you can see how the image is perceived by a robot similar to those that moderate content on social networks.
Here are a few examples.
So, we are going to analyze a creative promoting joint pain medication. The objects detected by AI belong to such categories as Animal and Food. Object recognition isn’t always correct.
The list of labels is much more detailed and accurate (except for the first tag).
The text has been recognized successfully.
As for SafeSearch, this feature uses five categories to detect explicit content and has returned the following likelihoods that each of these is present in the image:
If you want to reach new GEOs and audiences, maybe it is all waiting for you on Telegram? We’ve prepared some material about Telegram audiences. What are the messenger’s users like this year? How old they are, what they do, and what they are interested in!
As you can assume, certain moderation algorithms will arrive at the same conclusions.
The accuracy of face detection is quite good:
AMAZON Rekognition
The next tool is Amazon Rekognition. To start analyzing your image, you will need to sign up first.
So, here’s what Rekognition can tell you about your creatives.
As you may notice, the accuracy of object detection is not that high. According to Rekognition, the image displays a jaw rather than a joint.
As for facial analysis, we learned that the image below shows a face that appears to be male and that this man is not wearing glasses.
Amazon Rekognition also allows you to compare faces (which is probably useless when it comes to analyzing creatives but can certainly help you find out which of your parents you take after). 
However, this solution was able to correctly detect an English-language text (Vision failed at this task):
As for inappropriate content, Rekognition was only able to detect nudity.
Why should I care?
The testing process involves controlled modification of specific parameters and evaluating whether this optimization has been successful. To better understand what it is that you are changing, you need to clearly see the overall picture. As a human, you probably can’t break down your creative into smaller elements and be completely impartial. Fortunately, you have a few AI recognition tools at your disposal.
Each tool has its strengths and weaknesses, and so it’s up to you to decide which one to use. Obviously, artificial intelligence can’t help you figure out what is going to convert and what will fail to generate decent results. However, once you test out and analyze a slew of creatives, you will be able to take more educated decisions based on numbers rather than opinions.
There are a lot of shoes here apparently
