April 11th, files reveal that software giant Adobe has initiated the procurement of video assets to develop artificial intelligence video generation tools, in an effort to catch up with OpenAI's showcased text-to-video technology, Sora.
Adobe is compensating its community of photographers and artists $120 to gather videos depicting everyday activities such as walking, expressing joy, or anger, to train artificial intelligence.
Over the past year, Adobe has been integrating generative artificial intelligence technology into its flagship products Photoshop and Illustrator. Presently, the company has launched tools capable of generating images and illustrations from text, with usage reaching billions of times.
However, OpenAI's recent demonstration of the video generation model, Sora, has reignited concerns among investors that Adobe's long-standing dominance in the creative software industry could be disrupted by new technology. Adobe states that it is also developing video generation technology and plans to disclose more details later this year.
Adobe requests photographers and artists to submit over 100 short videos showcasing human actions and emotional expressions, as well as close-ups of body parts like feet, hands, or eyes. Simultaneously, the company also seeks videos showing people "interacting with smartphones or fitness equipment," emphasizing that submitted materials must not contain copyrighted content, nudity, or any "offensive material."
Compensation for submitted videos varies depending on their length, averaging around $2.62 per minute, with a maximum of $7.25 per minute.
A spokesperson for Adobe stated that company executives have made it clear that Adobe is developing video generation capabilities.
This move underscores the demand for large volumes of data in developing artificial intelligence models. Currently, there is widespread controversy surrounding the sources of this data. Mira Murati, Chief Technology Officer of OpenAI, mentioned in an interview last month that she herself is unsure whether Sora utilizes user-generated videos from Google's YouTube, Meta's Facebook, and Instagram for training.
Adobe primarily gathers training data through its vast marketing and creative agency media libraries to train models with uniqueness. In cases where stock content is not found, Adobe opts to directly purchase assets. For contributors who submit large volumes of photos for AI training, Adobe offers compensation ranging from 6 cents to 16 cents per image.
