Skip to content
IFAI

Generative media

Adobe pays $3 a minute for videos to train its AI

Published Reach 39

Adobe is paying creators an average of $2.62 to $7.25 per minute for short clips of everyday human actions and emotions to train its upcoming text-to-video generator. The data-buying push is aimed directly at countering OpenAI’s Sora, doubling down on Adobe’s pitch of commercially clean, copyright-safe training data.

Adobe pays $3 a minute for videos to train its AI
Image: nairametrics.com

While OpenAI danced around questions about where Sora got its training data, Adobe decided to pull out its checkbook.

According to a report from Bloomberg, Adobe is now offering to buy video clips from its contributor network of photographers and artists to train its upcoming text-to-video AI. The payouts average $2.62 per minute of footage, with some rates scaling as high as $7.25 per minute.

Buying human emotions by the minute

Adobe isn't just taking any footage off the shelf. The creative giant has specific gaps it wants to fill.

Contributors are being asked to submit batches of over 100 short clips showing everyday human actions and basic emotions, like walking, smiling, or looking angry. They are also seeking close-ups of human anatomy and interactions with everyday items like smartphones and gym gear.

These happen to be the exact areas where current video models struggle most: fingers folding awkwardly, fake-looking facial expressions, and bad object physics. Contributors are explicitly warned not to submit copyrighted or offensive material.

The commercially safe counterattack

After OpenAI dropped Sora earlier this year, nervous investors began questioning whether generative video would make Adobe's creative suite obsolete. Adobe promised it would reveal its own video tools later this year, and this buying spree is the groundwork for that promise.

It also highlights a massive clash of philosophies. When asked by The Wall Street Journal if Sora trained on YouTube or Instagram, OpenAI CTO Mira Murati gave vague, viral non-answers. Adobe, meanwhile, has doubled down on being the "commercially safe" alternative, relying on its vast stock library and paid submissions so corporate clients don't get sued.

Why it matters

If you are a photographer or video creator, this is an easy way to pocket a few hundred bucks by filming mundane daily tasks. But don't expect it to replace full-time creative work; it looks very much like a temporary bounty program for bulk training data.

For enterprise customers, Adobe's move is a breath of fresh air. Fortune 500 brands are terrified of copyright backlash, meaning many won't touch Sora with a ten-foot pole for commercial campaigns. Even if Adobe's video generator launches with slightly lower fidelity than Sora, the legal peace of mind alone guarantees it enterprise market share.

The catch? Paying $3 a minute for data does not scale nearly as easily as scraping the open web. Adobe's clean-hands approach is undeniably ethical, but we think it remains to be seen whether a curated diet of stock footage can ever match the raw power of models trained on the wild chaos of the internet.

Sources (1)

  1. [1]
    nairametrics.com(opens in a new tab)Originalnairametrics.com

Topic trackerAdobe's Generative Video AI Development

Every story on this topic, newest first.

  1. Adobe pays $3 a minute for videos to train its AIThis story
  2. Adobe creates 50-person CAVA research lab for video and audio AI

Best of AI, daily