"Hannah Alsark says that Adobe's top clients-the owners of some of the most protected, most valuable brands and IP in the world-had a stark message for the company regarding Firefly, its generative AI engine: They wanted more, and they wanted better. "They told us they actually needed models that understood all their products, all their brands, their creative direction," says Alsark, Adobe's VP of GenAI New Business Ventures."
"Firefly-which uses prompts to create assets across all Adobe's vector, bitmap, and motion apps-couldn't do this because it doesn't understand brands at the IP level. "We consider that a feature, not a bug," Alsark tells me. Firefly is a generic engine, but Adobe's top clients need to create millions of assets for dozens of different platforms and marketing media, all of them conforming to their own strict IP rules and brand books."
"A company with just eight products that wants to market them across 15 channels in 35 languages with a few refreshes a year is already looking at creating half a million individual assets. "With social, we know we're doing probably three refreshes a week," she explains. "So the real numbers are in the millions and millions and millions.""
Adobe created Adobe AI Foundry, a consultancy and model-building arm to deliver bespoke generative AI models trained on clients' IP and creative guidelines. Top clients require models that understand products, brands, characters, and motion styles to produce compliant assets at massive scale. Firefly remains a generic engine lacking IP-level brand comprehension, which firms view as intentional for general use. Major brands face a combinatorics math problem: even modest product sets multiplied across channels, languages, and refresh cycles yield millions of required assets. Adobe positions responsible, custom AI as the unlock to generate hundreds of thousands to millions of brand-specific images, illustrations, and videos.
Read at Fast Company
Unable to calculate read time
Collection
[
|
...
]