Adobe announced today a foundry-type service that builds unique generative AI models for big business — or what Adobe calls its “enterprise” clients.
Adobe’s new AI Foundry expands on the company’s Firefly family of models, offering specially trained models for enterprise clients as well as usage-based costs.
This is an evolution of Adobe’s business model away from “seat licenses” and towards value created.
At its core, this service refines Firefly — which are models trained on licensed and rights-cleared data — with one of a business’ own archives, product subtlety, or style guides.
Adobe notes that enterprise consumers have created over 25 billion assets with Firefly since launch in 2023, and Foundry takes that business and turns it into a bespoke system that creates text, photographs, videos, and even 3D scenes accurate to a brand.
What Adobe’s AI Foundry Offers Enterprise Clients
Foundry is a co-development initiative: Adobe collaborates with clients to develop training systems, describe brand limits, and establish tone and visual identity.
The ensuing models produce rule enforcers — logos in preventative areas, endorsed palettes, judicial advisers — so that all production complies with brand governance absent a designer monitoring each image.
Outputs may pass seamlessly into Creative Cloud apps including Photoshop, Illustrator, and Premiere Pro, as well as Adobe Express for light editing, and into enterprise schemes through APIs and Experience Cloud.
A product hero picture produced for one campaign may be immediately converted into dozens of configurations and platforms without breaching any brand message.
Why Enterprises Want Custom Generative AI Models
Global marketing teams are noosed by the imperative to version content across seasons, languages, and platforms — a work that’s repetitive, time-consuming, and slow. Generative AI can manage the scale, while humans attend to concept and craft. Analysts at Gartner have predicted that most businesses will take up such generative AI through APIs, or applications, in the next few years; and a report by McKinsey found it could add $2.6 trillion to $4.4 trillion in economic value each year, with marketing and sales among those capturing some of the biggest gains.
Control is the promise of custom models: instead of depending on general-purpose systems trained on the open web, a brand can limit outputs to sanctioned knowledge and visuals and voice. That’s especially important for regulated industries, where an errant claim or off-brand visual can represent real risk.
Branding and IP governance for enterprise AI models
Firefly’s training on licensed sources is a cornerstone of Adobe’s enterprise pitch. The company has been a proponent of the Content Authenticity Initiative and backed the C2PA manifest to represent content attributes, or metadata that can tell you who authored an asset and how it was changed. Look for Foundry deployments to make heavy use of those credentials to help businesses monitor AI-driven content and comply with long-promised disclosure norms from industry groups.
Companies also look for data isolation and audit trails. Adobe pitches Foundry as a way to keep customer training data siloed with permissions and review workflows so sensitive assets don’t leak across teams or clients. And for extra peace of mind, Adobe has historically provided IP indemnification for Firefly output in enterprise plans — which policy will be closely read as customers start scaling custom models.
Pricing and access for Adobe’s enterprise AI Foundry
Unlike traditional seat licenses, Foundry leverages consumption-based pricing, a structure that dovetails with production workloads — large during campaign bursts, small between product cycles — and allows finance teams to connect AI costs to measurable outcomes, like asset volume, localization breadth, and time to market.
Governance features and SLAs will be packed into enterprise agreements, and capacity planning and model refresh schedules will be based on a brand planning calendar and rollout.
The big cloud vendors already have model customization capabilities via platforms like Google Vertex AI, AWS Bedrock, and Azure OpenAI Service, and top-tier model firms provide fine-tuning and retrieval augmentation.
The distinction for Adobe is the depth of creative workflows and their demonstrated training prowess, rooted in licensed media, alongside direct connections to Creative Cloud and Experience Cloud. Getty Images also announced a similar agreement, and it is indicative of the crucial importance of provenance for enterprise approval.
Decisions will be based on whether design and marketing software integration is more important than cloud-agnostic ML stacks. Many will pick the latter, utilizing Adobe’s ecosystem for creative generation while relying on their cloud for domain-specific models.
A global retailer can create a master campaign at once and land it in hundreds of localized assets — copy in multiple languages, visuals adapted for climate and culture, and platform-specific formats — kept on the safe side of the legal line and in the proper side of brand tone.
A consumer electronics company could feed its product library to the model so that feature pull-outs, textures, and lighting result in the exact same look across launches and in each launch.
In media and entertainment, previsualization teams can spin up 3D scene variations that match a franchise’s visual bible, accelerating look development without replacing final artistry. Financial services and healthcare marketers can enforce compliance guardrails that block unapproved claims while still benefiting from speed and scale.
The winners in this next phase of generative AI will earn it with provenance, permissioned data, and human-in-the-loop design — not just bigger models.
Adobe is betting brand-safe scale is the feature enterprises value most. Adobe’s AI Foundry is less about replacing creatives and more about industrializing the content supply chain — codifying brand knowledge into a controllable model that delivers at scale. If it can prove trust, governance, and measurable ROI, enterprises will treat custom models as core infrastructure rather than experiments.
