Key Takeaways:
To catch up to AI-powered video generators from Google and OpenAI, graphics and illustration software developer Adobe is working with photographers and artists to supply content that will be used to train the company’s artificial intelligence models.
Adobe is Paying Creators for Images and Videos to Train AI Models
According to a Bloomberg report, Adobe is purchasing videos that show people engaging in various everyday activities like walking, running, using a mobile phone, working out, and others to show human emotions. The company also requested simple videos featuring human anatomy like feet, eyes, ears, and hands, an area where AI models struggle with depiction.
Adobe is paying between $0.06 and $0.16 for each photo and an average of $2.62 per minute for videos. The content will be used to help develop the company’s AI text-to-video generator that will join its existing lineup of image and illustration generator tools.
The media software provider will largely be using content from its library of stock media, but is buying content from creators to fill in any gaps in the materials it has available, reported Bloomberg on Friday, April 12th.
Adobe promised to reveal more information about its AI video-generating technology later this year. The company has been rolling out generative AI features across its digital media, publishing, and advertising products and plans to add more experiences in the coming months.
Adobe’s Text-to-Graphics Model FireFly Used by Major Players for Media Content
In 2023, Adobe released Firefly, a creative generative AI model that produces images, social media posts, flyers, vector graphics, and more from simple text prompts. Since its release, the model has helped users generate more than 6.5 billion images, vectors, designs, and text effects.
Speaking on its AI model, Adobe’s president of digital media business, David Wadhwani, said early adopters like IBM are putting Firefly at the forefront of their content creation processes. IBM used Adobe Firefly to generate more than 200 campaign assets and 1,000 marketing variations “in moments rather than months”.
AI Firms Face Scrutiny Over Unpermissioned Use of Licensed Content When Training Models
However, the company’s program of paying contributors for images and videos comes at a time when other firms are facing controversy over how they have used media content to develop their respective AI models.
Microsoft-backed AI startup OpenAI is at the center of the scandal for its use of diverse web content in training artificial intelligence products like ChatGPT, text-to-image model DALL-E, and now the text-to-video generator Sora.
OpenAI has also been sued by various authors and publications such as The New York Times over its use of licensed content without permission in training.
Earlier this month, YouTube CEO Neal Mohan told Bloomberg that while there is no concrete evidence to suggest that OpenAI has used videos from the platform to enhance the capabilities of its video-generation software Sora, any such use without seeking permission would be a violation of YouTube’s terms of service.
A Reuters article on companies purchasing media content to train AI models reported average prices of between $0.05 and $1 per photo and more than $1 per video. Stock image site Freepik licensed its more than 200 million image archives between $.02 and $0.04 per picture.
Meanwhile, tech giants like Google, Meta, Apple, and Amazon are paying up to $2 per image, between $2 and $4 per short-form video, and $100 to $300 per hour for longer videos.
More News: Apple Issues Mercenary Spyware Alert To iPhone Users In Over 90 Countries