Hacker News Reader

Comment by michael-online | Hacker News Reader

All the incentives are there, not too surprising. What percentage of people will notice? How long until only a trained eye will recognize it?

View on HN

Why not just grab a pic from the actual film?

If youtube is to be believed, thumbnails make a dramatic difference to first-glance engagement. A model trained to find great(by what metrics?) screenshots from the film sounds like a fun project.

https://recosense.medium.com/how-netflix-uses-ai-and-ml-to-d...

This reminded me of this article I read a few years back. I wonder if they still use the same method.

Netflix has written about their approach to solving that problem:

https://netflixtechblog.com/ava-the-art-and-science-of-image...

https://netflixtechblog.com/artwork-personalization-c589f074...

Netflix's approach has also had criticisms

https://www.theguardian.com/media/2018/oct/20/netflix-film-b...

I'm looking forward to classic movies with AI-generated thumbnails that take the main actor or actress and make them do the "Youtube Thumbnail Surprised Face". Imagine The Maltese Falcon or Citizen Kane but with thumbnails that look like [1].

1: https://hard-drive.net/wp-content/uploads/2021/05/youtubers-...

The first image you conjured in my mind was Indiana Jones holding the holy grial and doing a pog face.

This is the baffling part. Maybe it's like a "soft launch" thing to see if it works at all?

That would require paying a contractor for ~5 minutes of time to skip through the movie until they find a good scene, take the screenshot, trim it, and upload it. Ugh. We've already got all these GPUs over in AWS, just spin up an image generation model and prompt it "Make a thumbnail preview image for the film {{film.name}}", good enough.

That'd take a human with taste a terrifyingly large 5 minutes, whereas AI trash can be entirely programmatic, with maybe an intern picking the best of 4 in a couple seconds. Think of the savings, and with only a moderate impact on the reputation of your service.

Good point. I wonder if Amazon is experimenting with this with the idea that they will be able to make more "eye grabbing" thumbnails with genAI as opposed to pure stills from the given film. The click through rate may be a key metric internally, and since prime video is an add-on to prime the risk of customers dropping the service over this is low (compared to if Netflix did this)

Licensing fees. Why pay for individual images when you can just generate them and pay once for the tool.

What percentage of people will notice the primary means of marketing the film?

Almost all of them, surely. Maybe not consciously... but unconsciously certainly.

Films are incredibly expensive products to produce, an average of $100 million dollars apparently [1]. Do the incentives really not align to pay for professional artists to create a high quality thumbnail? I doubt it. Hell, youtubers have figured out it's worth doing this on every video which cost and pull in 4-5 orders of magnitude less money.

[1] https://www.investopedia.com/financial-edge/0611/why-movies-...

I mean, it's amazing how much we already put up with terrible Photoshops on movie covers and thumbnails. This is really just the next evolution.

Reply on news.ycombinator.com