Source: Financial News Agency Author: Liu Rui

After OpenAI's Wensheng video model Sora attracted enthusiastic attention in the market, Adobe couldn't stand it either. Currently, the company has sped up the pace and began purchasing videos to build its AI Wensheng video model.

It is worth mentioning that OpenAI is currently facing controversy over the source of its AI model training data, and Adobe's practice of purchasing videos directly from photographers and artists will obviously be considered more compliant.

Adobe is procuring video resources to train AI models

According to a document, the software company is currently paying its network of photographers and artists $120 each to submit videos of people's daily activities, such as walking or expressing emotions of joy and anger. The company wrote that the goal is to provide resources for AI training.

Over the past year, Adobe has been working to add generative artificial intelligence capabilities to its software portfolio for creative professionals, including Photoshop and Illustrator. The company has released tools that use text to generate images and illustrations, and these tools have been used billions of times so far.

Despite this, after OpenAI launched a demonstration of its Wensheng video model Sora, Adobe investors became increasingly concerned that the company, which has long been a leader in the creative software industry, could be disrupted by this new technology.

To this end, Adobe said it is speeding up the development of Wensheng video technology and plans to discuss more related content later this year.

Adobe requires photographers and artists to provide more than 100 short videos of people's movements and emotions, as well as simple anatomical photographs of feet, hands, or eyes. The company also wants photographers to shoot videos of “people interacting with items such as smartphones or fitness equipment.” The average cost to submit a video is around $2.62 per minute, but it can also be as high as $7.25 per minute.

It also warned against providing copyrighted material, nudity, or other “offensive content.”

A more compliant source of data?

It's worth mentioning that this list from Adobe highlights the vast amount of data needed to build AI models. And Adobe's most direct competitor, OpenAI, is currently facing controversy over these data sources.

OpenAI Chief Technology Officer Mira Murati (Mira Murati) said in an interview last month that she is unsure whether Sora has received user-generated video training on Google's YouTube and Facebook and Instagram under Meta Platforms Inc.

Murati's flickering remarks have raised questions from the industry about the source of OpenAI training data. YouTube CEO Neal Mohan publicly stated last week that although he has no direct evidence that OpenAI used YouTube videos to train Sora, if this is true, then OpenAI “clearly violates” YouTube's terms of use.

By contrast, the AI model training data source Adobe is seeking is clearly more secure: Adobe mainly uses its vast media library for marketers and creative agencies. In the event that its inventory library is insufficient, it takes images directly from contributors.

In addition to video resources, it also rewards contributors who submit numerous photos for AI training. According to recruitment information, these photos used for AI training are paid between 6 cents and 16 cents per image.

edit/lambor