You understand that we Sapiens are visual animals, yes?
You and I can move our eyes 3-4x per second to process information.
And we analyze images in as little as 13 milliseconds.
In 2014, Netflix consumer research showed that their thumbnail artwork:
- Influences people the most with content selection and watching it.
- Is the focus 82% of the time when browsing Netflix library.
You’d only look for 1.8 seconds and if you can’t find content to your liking, you’re likely to exit the app.
This is why Netflix uses an elaborate selection of thumbnails for each of its 200m+ users.
Aesthetic Visual Analysis (AVA)
AVA starts by pulling all the frames from a video.
Frame Annotation process the tag frames with metadata identifying variables. Key variables include:
- Frame no.
- Brightness or contrast
- Nudity probability
- Face or skin tone
Then they grade frames based on:
Visual variable includes brightness, contrast, color, motion blur, and so on.
Contextual variable comprises face detection and shot angle.
Composition variable weighs on photography principles, like “rule of thirds”, symmetry, depth of field, etc.
The image ranking step chooses the best thumbnails that are likely to get clicks.
A winning combination revolves around:
- Expressive faces
- Main characters
Another trait that sets aside Netflix from the competition is good localization. Netflix displays thumbnails for a specific region or country which the locals find most attractive.
As per Netflix, thumbnails with bad guys in them outperform. Also, thumbnails with over 3 people in them vastly underperform.
Thumbnails: More than three characters vs single character
In their research, Netflix found people with more than 3 characters vastly underperform when compared to thumbnails of one character.
The reason Netflix invests too highly in artwork optimization is that those provided by studios don’t cater to their streaming app.
The creative work Netflix received was best suited for DVD covers or billboards, as seen in the example below.
Machine Learning and A/B testing
How does Netflix figure out which thumbnail to show me?
It does so by relying on ML or Machine Learning. For example, based on your recent watch history, if you’ve viewed rom-com, you get the following thumbnail for “Good Will Hunting” and if you’ve watched comedy, you get the one below it.
Similarly, Uma Thurman fans get the first “Pulp Fiction” thumbnail. And John Travolta fans see the second one.
This shouldn’t come as a surprise, but Netflix engages in A/B testing. Which thumbnails to show its users. That’s why the artwork constantly changes.
However, all this isn’t without some heat. In 2018, they accused Netflix of creating artwork based on race. “Like Father” which has mainly Caucasian cast, a Black user saw the thumbnail on the right when scrolling through titles.
Netflix disputed the claim, saying they target based on viewing history instead of demographics.
Harmless case of thumbnail optimization
A slick method to optimize thumbnails and props to Netflix for showing Anya Taylor-Joy on the cover of Peaky Blinders instead of Cillian Murphy or other notable cast members. The female star’s “Queen’s Gambit” (Netflix’s very own) skyrocketed last year during the pandemic and Netflix banked on its popularity serving thumbnails of her, when she’s only there in 6 out 30 episodes.
I’m sure it drove mad clicks! You can’t help but love Netflix thumbnail algorithm.
The only exception to artwork “rules” is Squid Game. For now, I’ve only seen 3.
Imagine Netflix home page without thumbnail artwork.