Adobe is reportedly developing an AI-powered text-to-video generation model and plans to enhance its capabilities by sourcing video content from photographers and artists. The company intends to use this content, along with existing libraries of stock images and videos, to efficiently train its AI models. According to Bloomberg citing sources, Adobe offers an average compensation of $2.62 (approximately Rs. 220) per minute for submitted video content.

What are Adobe’s plans?

Adobe has reached out to its network of photographers and artists, offering to pay up to $120 (approximately Rs. 10,000) for submitting videos. The videos were asked to show individuals engaging in everyday activities and expressing a range of emotions, including joy and anger. Adobe’s goal appears to be to use this data to train its artificial intelligence models to recognize and replicate human expressions and natural movements.

Additionally, Adobe requires certain types of video content, such as short clips depicting people expressing emotions and footage focusing on various aspects of human anatomy, such as hands, feet, and eyes. Additionally, the company looks for videos that show individuals interacting with objects such as smartphones and fitness equipment.

Documents obtained by Bloomberg stress the importance of adhering to guidelines regarding copyrighted material, nudity and offensive content when submitting videos. While the average compensation offered is $2.62 per minute, some contributors may receive up to $7.25 (approximately Rs. 600) for their submissions.

The rise of AI models

The development sheds light on a growing trend of companies investing in data procurement to train artificial intelligence models as public data sources increasingly dry up. While some companies choose ethical data acquisition methods, others face criticism for allegedly using copyrighted material from social media platforms. For example, a recent report claimed that OpenAI used more than 1 million hours of YouTube video transcription data to train its GPT-4 model.