Each enterprise team may require a different level of AI support. Selecting a model is an ongoing decision shaped by project goals, governance requirements and production readiness. It’s not about adopting a more advanced model, but rather it is about matching the model to the organisation's needs when it comes to managing brand imagery and protecting proprietary IP rights. A team generating campaign imagery with minimal governance requirements may need something very different from one managing global campaigns, regional adaptations and established brand standards. As content operations become more complex, organisations often require greater model specificity to support them.
Let's look at the three stages of customisation across Firefly:
Stage 1: Adobe Firefly base models
Many organisations start using generative AI models for specific creative tasks rather than redesigning their content operations. Adobe Firefly base models provide a practical starting point for those teams by bringing AI into creative workflows. Teams can use them to develop concepts, build mood boards, explore campaign directions, create early content variations and accelerate ideation.
At this stage, the priority is often to learn where AI fits within existing ways of working and to identify opportunities to create value. For example, Gatorade used Adobe Firefly for enterprise to power personalisation experience through its signature bottles on its ecommerce platform. By generating hundreds of thousands of unique designs, the brand expanded creative possibilities for consumers while increasing engagement and strengthening brand loyalty. Experiences like this give teams hands-on knowledge of AI-assisted asset creation while building the skills and processes needed for broader adoption.
Governance at this stage starts with setting the right foundation. Organisations define usage guidelines, apply human review, document prompt best practices and establish approved workflows for commercially safe experimentation. This equips teams to explore AI capabilities while preparing for broader adoption over time.
As content production scales, teams need more than generic AI-generated images. They want images that match their brand.
Stage 2: Adobe Firefly Custom Models
Brand consistency becomes more difficult to maintain as the volume of AI-generated content increases. Marketing and creative teams need their AI-generated images to visually reflect specific characters, illustration styles, campaign aesthetics and photographic styles. Adobe Firefly Custom Models support these requirements by training on a small, curated set of approved brand assets, such as icons, character libraries, illustrations and campaign assets.
Firefly Custom Models can be trained on dedicated assets for a specific campaign, such as a particular character or photoshoot style in a single model. By learning visual patterns, the models generate more consistent, on-brand output. They are great for teams that have small training sets — just 10 to 20 assets — and want to train and use the model quickly, accelerating content production. With Firefly Custom Models, teams can explore variations, accelerate production and move from concept to final assets more efficiently.
For example, Amazon Fresh partnered with KINESSO to implement Adobe Firefly Custom Models trained on campaign assets. The team adapted creative assets from one seasonal campaign to another, producing high-quality holiday-themed visuals and cut production time by 93%, all while maintaining brand consistency and commercial safety. Enterprises looking to localise on-brand visuals across audiences, markets and channels can build a custom model and reduce time-consuming product reshoots.
Governance becomes more structured at this stage. Organisations train models on approved brand assets, control access to custom models and validate outputs before scaling production. These practices maintain consistency while enabling creative teams to keep up with content demands.
Stage 3: Adobe Firefly Foundry
Leading brands and franchises rarely create content around a single product, character or campaign. They manage portfolios of assets, experiences and IP that span channels, formats and customer touchpoints. At this stage, teams need custom models that can preserve relationships across those systems while supporting the creation of image, video, audio, vector and 3D content.
Adobe Firefly Foundry is a managed enterprise model-development product for organisations with complex IP and multimodal production needs. Firefly Foundry employs large-scale training to teach the model multiple brand concepts across formats — images, audio, video and more — and tie the relationships between them. Organisations can build private models trained on proprietary assets and retain ownership of both the training data and model outputs. The aim is to combine creative speed with creative precision to support production-ready AI at an enterprise scale.
With this approach, the model aligns with broader brand and franchise systems across content workflows. For example, Disney uses Firefly Foundry to help Imagineers quickly generate consistent visual iterations across the characters, stories and attractions that audiences connect with and love. As Disney expands across parks, hotels, cruises and other experiences, proprietary models help its teams meet growing design demands while staying true to the brand’s IP.
Governance requirements also expand at this stage. Organisations are required to manage proprietary IP, make model deployment decisions, integrate models into production workflows, establish stakeholder oversight and implement enterprise-level controls for production use. They also need AI usage guidelines for employees, vendor evaluation processes and governance practises that evolve along with AI adoption.
