Hoping to draw more enterprise teams to its ecosystem, Adobe launched a brand new model customization service called Adobe AI Foundry that will create customized versions of its flagship AI model, Firefly.
Adobe AI Foundry will work with enterprise customers to revamp and retrain the architecture Firefly models specific to the shopper. AI Foundry version models differ from Firefly custom models in that Foundry models understand multiple concepts, in comparison with custom models that only understand a single concept. These models will even be multimodaland provides a broader use case than custom Firefly models that may only capture and respond to photographs.
Adobe AI Foundry models powered by Firefly understand an organization's brand tone, image and video style, services and products, and all mental property. Based on this information, the models generate content for every use case desired by the corporate.
Hannah Elsakr, vice chairman of GenAI New Business Ventures at Adobe, told VentureBeat that the thought to create AI Foundry got here about because enterprise customers wanted more sophisticated custom versions of Firefly. But because enterprise needs are so complex, Adobe will redesign the architecture reasonably than handing the reins to customers.
“We will retrain our own commercially secure Firefly models using the company IP. We keep this IP separate. We never incorporate it into the bottom model again, and the corporate itself owns this output,” Elsakr said.
Adobe will provide the foundry version of Firefly through its Firefly Services API solution.
Elsakr likened AI Foundry to a consulting service because Adobe can have teams working directly with enterprise customers to retrain the model.
Deep tuning
Elsakr calls Foundry a deep tuning method since it goes beyond just fine-tuning a model.
“We're pondering, perhaps more in layman's terms, that we surgically reopen the Firefly-based models,” Elsakr said. “So they profit from all of the knowledge on the planet from our image model or a video model. We return in time and convey in the corporate's mental property, like a brand. It may very well be footage from a shooting style, regardless of what license they're allowed to contribute. Then we retrain. We call this continuous pre-training, where we chubby the model to set some things in a different way. So we're literally training our base model latest and due to this fact call it “Deep Tuning” as a substitute of “Deep Tuning”. Fine-tune.”
Part of the training pipeline involves Adobe's embedded teams working with the corporate to discover the information it needs. The data is then securely transmitted and captured before being tagged. It is fed to the bottom model after which Adobe begins a pre-training model run.
Elsakr claims that the foundry versions of Firefly won’t be small or distilled models. Often the extra data from corporations expands Firefly's parameters.
Two early customers of Adobe AI Foundry are Home Depot and Walt Disney Imagineering, Disney's research and development arm for its theme parks.
“We are continually on the lookout for modern ways to enhance customer experiences and optimize our creative workflows. Adobe's AI Foundry represents an exciting step forward in introducing cutting-edge technologies to deepen customer engagement and deliver impactful content across our digital channels,” said Molly Battin, senior vice chairman and chief marketing officer at The Home Depot.
More customization
Companies often turn to Fine-tuning and model adjustment to bring large language models with their extensive external knowledge closer to the needs of their company. Fine-tuning also allows business users to make use of models only within the context of their organization's data, so the model doesn't respond with text that has nothing to do with the business.
However, most organizations do the fine-tuning themselves. They hook up with the model's API and start retraining to receive answers based on their ground truth or preferences. There are several methods of fine-tuning, including some that will be done with only one request. Other model providers are also attempting to make it easier for his or her customers to fine-tune models, e.g OpenAI together with his o4 mini argumentation model.
Elsakr expects that some corporations can have three versions of Firefly: the foundry version for many projects, a custom Firefly for specific single-concept use cases, and the bottom Firefly because some teams desire a model that’s less burdened by corporate knowledge.
