If you are regularly uploading videos to YouTube, you have experienced some pretty terrible thumbnail choices the platform has historically provided. That is why we create separate thumbnails for our clients. Having a good-looking thumbnail can make or break someone deciding to watch your video (which is why a lot of sneaky folks like to add pretty ladies in their thumbnails).
In the past, YouTube basically grabbed frames from the video at the ¼, ½ and ¾ mark of the video. There were a few hacks you could use to work around getting an out-of-focus or garbled image as your thumbnail. Then a few years ago, YouTube allowed just about anyone to upload their own custom thumbnails.
Well, the platform announced this week an update to its thumbnail generator that promises to result in better video thumbnails. It does this by grabbing one frame per second while the video is being uploaded and then evaluates those frames as either good or bad using an algorithm along with what it calls deep neural networks (DNNs) in computer vision.
From its testing, YouTube reports that the new images are preferred 65 percent more than the previous thumbnail options. You can read their blog post, Improving YouTube video thumbnails with deep neural nets.
You can also read our blog post on tips for creating your own great thumbnails.
While this is an overall improvement to what was currently available to most YouTube uploaders, our recommendation is to continue uploading your own custom thumbnails that can also include your brand logo.
And for those of you who suspect Google ulterior motives, the updated DNNs and associated algorithms suggest the web giant is closer to watching and understanding what goes in the video.