Content
summary Summary

Photobucket and other internet veterans are now making money from their old data archives by licensing them to technology companies to train AI models, Reuters reports.

According to industry insiders, major technology companies such as Google, Meta, Microsoft, and OpenAI are currently licensing images, videos, and other content on a large scale to train their AI models. They pay anywhere from 5 cents to $1 per photo and more than $1 per video, depending on the buyer and the type of material.

Prices can vary widely: Defined.ai, which licenses data to Google, Meta, Apple, Amazon, and Microsoft, quotes prices of $1 to $2 per image, $2 to $4 for short videos, and $100 to $300 per hour for longer films. Nude images, which require special handling, can cost $5 to $7 per image. Text costs $0.001 per word.

Photobucket CEO Ted Leonard is currently negotiating with several technology companies to license 13 billion photos and videos at prices ranging from 5 cents to $1 per photo and more than $1 per video. His company cannot keep up with the demand: One buyer has requested billions of videos.

Ad
Ad

Shutterstock has signed deals with Apple, Meta, Google and Amazon for the use of hundreds of millions of images, videos and music files. The initial deals were for $25 to $50 million, but have been expanded. According to Reuters, OpenAI was one of the first customers.

Spanish platform Freepik has licensed most of its 200 million image archive to two major technology companies for 2 to 4 cents per image. CEO Joaquin Cuenca Abela says five more similar deals are in the pipeline.

Is the sale of user data legitimate?

Photobucket CEO Ted Leonard believes he is on the safe side legally. He points to an October update to its terms of service that gives the company the "unrestricted right" to sell uploaded content to train AI systems. He sees licensing as an alternative to advertising to continue offering free accounts.

However, the US Federal Trade Commission (FTC) warned companies in February against retroactively changing terms of use for the use of AI.

It may be unfair or deceptive for a company to adopt more permissive data practices—for example, to start sharing consumers’ data with third parties or using that data for AI training—and to only inform consumers of this change through a surreptitious, retroactive amendment to its terms of service or privacy policy.

FTC

The agency is investigating a training data deal Reddit struck with Google. Reddit is considered one of the most important data providers for AI companies.

Ad
Join our community
Join the DECODER community on Discord, Reddit or Twitter - we can't wait to meet you.
Ad
Join our community
Join the DECODER community on Discord, Reddit or Twitter - we can't wait to meet you.
Recommendation
Support our independent, free-access reporting. Any contribution helps and secures our future. Support now:
Bank transfer
Summary
  • Photobucket and other big data archive providers license images, videos, and text to technology companies such as Google, Meta, and OpenAI to train their AI models.
  • Prices range from 5 cents to $7 per image, over $1 per video, and $0.001 per word, depending on the type of material and the buyer.
  • Ted Leonard, CEO of Photobucket, is currently negotiating with several companies to license 13 billion photos and videos. Shutterstock and Spanish platform Freepik have also signed lucrative deals with major technology companies. Demand seems to be outstripping supply.
Sources
Online journalist Matthias is the co-founder and publisher of THE DECODER. He believes that artificial intelligence will fundamentally change the relationship between humans and computers.
Join our community
Join the DECODER community on Discord, Reddit or Twitter - we can't wait to meet you.