Most marketers believe it It is very valuable to have relevant, engaging images in content.
However, choosing the “right” images for blog posts, social media posts, or video thumbnails has historically been a subjective process. Social media and SEO gurus have a range of advice on choosing the right images, but that advice is usually lacking in real empirical data.
This got me thinking: is there a data-driven ̵
The technology to find the best photos
In July 2019, a fascinating new machine learning paper called Intrinsic Image Popularity Assessment was published. This new model has found a reliable way to predict the likely “popularity” of an image (estimating the likelihood that the image will achieve a similar result on Instagram).
It also demonstrated the ability to outperform people with an accuracy of 76.65% in predicting how many likes an Instagram photo would get, versus a human accuracy of 72.40%.
I used the model and source code from this document to find out how marketers can improve their chances of choosing images that will have the best impact on their content.
Finding the best screen caps for a video
One of the most important aspects of video optimization is choosing the thumbnail of the video.
According to Google, 90% of the best performing videos on the platform use a custom selected image. The click rates and ultimately the number of views can be greatly influenced by how noticeable a video title and thumbnail are to a searcher.
In the past few years, Google has been applying AI to automate thumbnail extraction from videos. The aim was to help users find thumbnails from their videos that were more likely to attract attention and click-through rates.
Unfortunately, with only three options to choose from, it is unlikely that the thumbnails Google currently recommends are the best thumbnails for a given video.
This is where AI comes in.
With a simple code, it is possible to counteract the “intrinsic popularity rating” (derived from a model similar to that described in this article) all each frame in a video, which gives a much wider range of options.
You can find the code here. This script downloads a YouTube video, splits it into frames as JPG images, and runs the model on each image, providing a predicted popularity value for each frame image.
Reservation: It is important to remember that this model has been trained and tested on Instagram images. Given the similarity in behavior when clicking on an Instagram photo or YouTube thumbnail, we believe it is likely (although never tested) that a thumbnail that is likely to work well as an Instagram photo will work as well as a YouTube video Thumbnail.
Let’s look at an example of how this works.
We looked at the intrinsic popularity model at three frames per second of this 23 minute video. It took about 20 minutes. The following were my favorites out of the 20 images that had the highest total scores.