Hoping to attract more enterprise teams to its ecosystem, Adobe Adobe launched a new model customization service called AI Foundry, which will create bespoke versions of its flagship AI model, Firefly.
Adobe A Foundry will work with enterprise customers to re-search and re-train Firefly model Specific to the client. AI Foundry version models differ from custom Firefly models in that foundry models understand multiple concepts compared to custom models with only one concept. This model will do Be multimodal toooffer a wider use case than custom Firefly models, which can only consume and respond to images.
The Adobe AI Foundry model, with Firefly at its base, will understand a company’s brand tone, image and video style, products and services, and all of its IP. Models will generate content based on the information a company wants for any use case.
Hannah Elsker, vice president of Genie New Business Ventures at Adobe, told VentureBeat that the idea to set up AI Foundry came about because enterprise users wanted more sophisticated custom versions of Firefly. But given how complex the needs of businesses are, Adobe may be recontracting rather than handing over to customers.
“We will retrain our Firefly commercially secure models with enterprise IP. We keep that IP separate. We never feed it back into the base model, and the enterprise itself owns that output,” Elsaker said.
Adobe will deploy the Foundry version of Firefly through its API solution, Firefly Services.
Elsker likens AI Foundry to a consulting service, since Adobe teams will work directly with enterprise users to retrain models.
Deep tuning
Elsaker describes the foundry as a deep tuning method because it’s more than just fixing the model.
“The way we think about it, maybe in more layman’s terms, is that we’re surgically reopening firefly-based models,” Elsker said. “So you get the benefit of all the knowledge in the world from our image model or video model. We’re going back in time and taking IP from the enterprise, like a brand, it could be footage from a shot style, which they have a license to contribute, and then we call that pre-training, which is why we retrain certain things differently. Fine-tuning.”
Part of the training pipeline includes Adobe’s embedded teams that work with the company to identify their data. The data is then securely transmitted and ingested before being tagged. This is fed to the base model, and then a pre-training model run is started from Adobe.
Elsaker maintains the foundry version of the Firefly will not be a miniature or distilled model. Often, additional data from companies increases the parameters of Firefly.
Two early users of Adobe eFoundry are Home Depot and Walt Disney Imaging, Disney’s research and development arm for its theme parks.
“We’re always looking for innovative ways to enhance our customers’ experience and streamline our creative workflows. Adobe’s AI Foundry represents an exciting step in embracing the latest technologies to deepen customer engagement across our digital channels and deliver impactful content,” said Molly Button, senior vice president and chief marketing officer at Home Depot.
More customization
Enterprises often refer Fine tuning and model customization To bring large language models with their extensive external knowledge closer to their company’s needs. Fine-tuning also enables enterprise users to use models only in the context of their organization’s data, so the model doesn’t respond with text completely unrelated to that business.
However, most organizations fine-tune themselves. They connect to the model’s API and start retraining it to respond based on their ground truth or their preferences. There are several methods for fine tuning, some of which include: Just with a hint. Other model providers also try to improve models for their customers, eg Open Eye With that O4-Mini reasoning model.
Elsaker said he expects some companies to have three versions of Firefly: a Foundry version for most projects, a custom Firefly for specific single-concept use cases, and a base Firefly because some teams want a model that’s less driven by corporate knowledge.