X

Pornhub is adding AI to sort and label its videos

One of the top adult streaming sites is adding smarter AI to tag adult stars and sex positions.

Scott Stein Editor at Large
I started with CNET reviewing laptops in 2009. Now I explore wearable tech, VR/AR, tablets, gaming and future/emerging trends in our changing world. Other obsessions include magic, immersive theater, puzzles, board games, cooking, improv and the New York Jets. My background includes an MFA in theater which I apply to thinking about immersive experiences of the future.
Expertise VR and AR, gaming, metaverse technologies, wearable tech, tablets Credentials
  • Nearly 20 years writing about tech, and over a decade reviewing wearable tech, VR, and AR products and apps
Scott Stein
2 min read
ai-1
Stephen K. Woo

Pornhub wants to make it easier for you to find that special video.

The company, one of the top streaming adult video sites, is adding AI tools to recognize and tag faces and physical attributes (like "brunette" above) of performers in its video library. 

Face recognition will identify 10,000 actors in the company's database (but not amateurs) and tag videos for better cross-referencing. The AI will rely on user feedback to upvote or downvote the autotagged suggestions to improve the system over time. 

According to Pornhub VP Corey Price, the AI platform will also identify sexual positions, locations and performers' features, in addition to using these identifications to tag and improve categorization. Pornhub plans to scan its full collection of over 5 million videos over the next year.

ai-2
Enlarge Image
ai-2

What Pornhub plans to do with its videos.

Pornhub
Watch this: My conversation with Harmony the sexbot

Price answered a few of our questions to clarify.

Will this also be used for spotting things like copyright violations? 

Price: We already use digital fingerprinting technology provider similar to YouTube's ContentID to scan our videos.

Will it spot duplicates of videos and sort accordingly?

The technology has the potential to do something like this in the future.

Will only the adult stars in your database be categorized or will it try to cluster other faces, like Google Photos?

We are only matching with porn stars in our database.

How many things can it identify? Does it identify positions or motion? Can it be used to categorize in existing slots only, or can it help define new and emerging categories?

Right now, we are only identifying porn stars. So far we've used the model on about 500K featured videos (which includes user submitted) and we plan to scan the whole library in the beginning of 2018. Very shortly, the technology will also be used to detect various sex positions/categories and be able to properly tag them as well.