Adobe is reportedly paying for videos to train new Text-to-Video AI models

Adobe is reportedly working on building a text-to-video generative model that leverages artificial intelligence (AI). The company is said to be purchasing videos from photographers and artists to train its AI models. This data is used in addition to the platform's existing library of stock images and videos. Interestingly, the software giant pays an average of $2.62 (approximately Rs. 220) for every minute of video posted. Notably, earlier this year, Adobe announced Project Music GenAI Control, an AI music generation tool that is still in development.

The company will offer a network of photographers and artists up to $120 (approximately Rs. 10,000) to post a video, which will be used to train the company's AI video model, Bloomberg reported. Based on documents seen by the newspaper, the company added that it requested videos of “people performing everyday activities such as walking or expressing emotions such as joy or anger.” It appears Adobe may use this data to train human facial expressions and natural movements through these sourced videos.

Additionally, the report highlighted that the software giant requires as many as 100 short clips of people expressing emotions, as well as shots of human anatomy such as feet, hands, and eyes. It also wants videos of people “interacting with objects” such as smartphones or gym equipment.

The document, seen by the publication, warns contributors not to post copyrighted material, nudity or other offensive content. The average payout for this task is $2.62, but the report says that depending on the poster, he could be paid as much as $7.25 (approximately Rs. 600) for posting a one-minute video. … apparently …

This trend also highlights the increasing costs that companies are currently incurring to procure data to train AI models, as publicly available data sources are exhausted. . While some tech companies source data ethically, others have been accused of stealing copyrighted data from social media platforms. A recent report claims that OpenAI used over 1 million hours of transcription data from YouTube videos to train his GPT-4.