Adobe is building its own AI models that can convert text into video, and unlike other companies, it plans to actually pay the creators of the materials used for training.
Using AI to generate video from text looks like the next big challenge in AI content creation, and OpenAI is taking the lead with its Sora AI model. But whereas OpenAI often sourced its AI training materials from every public source possible (and was accused in a lawsuit of allegedly containing copyrighted works) , Adobe is reportedly going down a different path, purchasing and sourcing footage from certain companies to train its next generation AI models. Film producer.
The videos Adobe is requesting are supposed to show fairly mundane things like phone usage, people expressing emotions, and human body movements, with Bloomberg reporting that Adobe is purchasing 100 different types of footage. From what you've said, it's probably more than that. The report calculates that compensation for submitted footage ranges from $2.62 to $7.25 per minute, with each video valued at $120.
It is unclear who is eligible to submit this footage to Adobe and whether there are limits to the amount of footage that can be submitted. If one person can post at least 100 clips of her at $120 each, that's a sizable paycheck for filming everyday events.
Additionally, posts must adhere to at least some rules. Bloomberg says any footage can't be offensive, show nudity or be copyrighted material, which is Adobe's standard. When the company ran a similar program to train its Firefly AI image generator, it wouldn't accept photos of brands or celebrities, or even certain words, according to Bloomberg. Similar rules could apply to Adobe's program for training video generators.
However, copyright may be the biggest concern for Adobe. OpenAI has been embroiled in countless lawsuits over ChatGPT, which generates summaries of copyrighted novels because it was allegedly trained on copyrighted novels. In an interview with the Wall Street Journal last month, OpenAI's chief technology officer Mira Murati said she didn't even know if Sola was trained in the YouTube videos, which YouTube CEO Neil Mohan said it was a violation of the platform's terms of service.
Whether it is permissible to use copyrighted works to train AI models is still legally undecided. Additionally, while Adobe's plan may be expensive, it could allow the company to source higher-quality footage and tailor future video generation tools to Adobe's preferences. ®
