India’s digital art creators have an opportunity to monetise their skills by producing training data for generative AI image and video models, said Alexandru Costin, vice president of GenAI at Adobe.

Instead of fearing loss of work opportunities in the GenAI era, designers, photographers and illustrators are embracing the new technology. They are adding a new income stream for each digital asset they create for model training, he said. The company, which has built a proprietary family of text-to-image and text-to-video GenAI models known as Firefly, says it is running “content missions,” which identifies gaps in training data (images and videos) and rewards contributors who fulfil these missions.

“We put missions on Adobe Stock and tell contributors that we have a gap,” Costin said. “For example, if we don’t have photos of Indian culture of a select province, we put out a stock mission inviting photos and videos with this particular cultural characteristic.”

They participate in the marketplace in real time, and are paid per asset. “Since the launch of Adobe Firefly last year, we’ve seen an enthusiastic response, with Indian creators contributing towards the 13 billion plus image generations created with Firefly,” he said. “This has enriched Adobe Stock with culturally relevant assets for a global audience.”

India accounts for Adobe’s largest employee base outside the US with 8,000 people in five campuses. “Some of our teams are working on the Firefly model, others are operating our GPU (graphics processing unit) cluster while some of them are building services like the Adobe Illustrator,” he said.

Threat to jobs

On whether AI poses a threat to jobs such as stock contributors and photographers, Costin said the demand for digital content is going to increase exponentially.

“To me, GenAI is like the new digital literacy. And the only way to succeed is to embrace it. You need to learn how to level up yourself from a person that moves pixels on screen to a creative director that can guide GenAI to generate 100 times more content that you can sell,” he said. “Yes, creating content is now 100x easier with GenAI, but, at the same time, it is also creating an infinite need for content.”

Enterprises and consumers want personalised content. AI will enable a new level of personalisation where every user will see content that matches their expectations, aesthetic style, sentiments etc., he said. Adobe, the developer of tools like Illustrator, PhotoShop, PDF, Acrobat Reader and Flash, embarked on its GenAI journey in 2022 with the Firefly models meant for “safe commercial use,” Costin said. It filtered out unlicensed copyright images from training and used over 400 million assets (images, illustrations, videos) of the Adobe Stock Marketplace.

“So besides the normal revenue the stock contributors are getting from Adobe Stock, we also give them a yearly bonus, every time their data enters into the Firefly training dataset,” he said. “Our 2024 bonus is at an all-time high.”

Compute cost

The high cost of compute needed for training frontier image and video models is an area of concern, according to Costin.
“Fortunately, we have been doing a lot of GPU optimisation and moving into both large distributed training and efficient inference,” he said. “We are keeping our GPU utilisation very high, and managing our clusters, so we don’t keep GPUs idle, because that consumes unnecessary power.”

New techniques like quantisation are getting the model to operate at the same quality with much less data.