On November 9, 2026, Adobe and The Walt Disney Company announced a multiyear strategic partnership to jointly develop AI models tailored to Disney's characters and story worlds. Adobe becomes Disney's official AI innovation partner and will extend the theme park design collaboration announced in June to StudioLAB and ESPN. At Imagineering, Disney's theme park development arm, two tools are already in use: an AI model that generates 2D concept art from hand-drawn sketches, and an image generation model that reproduces the distinctive worlds of Disney's works. The partnership is not meant to open up Disney character generation to the general public. Its purpose is to support Disney's internal production work.
At the core of the collaboration is Adobe Firefly Foundry, Adobe's enterprise AI development service. It lets a company co-develop AI models tailored to its own assets and style, and build them into real production workflows, which goes beyond simply fine-tuning a model. The Disney collaboration shows which tasks generative AI is trying to make more efficient on the production floor, and what matters when judging the results.
An AI Partnership Expanding from Theme Park Design to StudioLAB and ESPN
The collaboration between Adobe and Disney Imagineering was not first announced this time. In an announcement on June 16, 2026, the two companies said they would develop AI using Disney's proprietary design assets. The use cases named at the time included generating 2D concept art from hand-drawn sketches, creating original imagery, and extending 2D images into 3D models. The November 9 announcement states explicitly that two of those image generation use cases are already in use, and the scope of joint development now reaches other parts of Disney.
The progress of the collaboration, broken down by division, use case, and stage of adoption, looks like this.
| Announcement date | Division | Announced use cases | Confirmed stage of adoption |
|---|---|---|---|
| June 16, 2026 | Imagineering | 2D concept art from hand-drawn sketches, original image creation, 2D-to-3D extension, and more | Collaboration and initial use cases announced. Not all planned features are live |
| November 9, 2026 | Imagineering R&D | 2D concept art from sketches; original images tailored to the worlds of Disney's works | Two use cases stated to be already in use |
| November 9, 2026 | StudioLAB | Prototyping and testing of internal production processes | Joint projects underway |
| November 9, 2026 | ESPN Edge | Deepening engagement with sports fans; using archival footage | Joint projects underway |
The partnership can therefore be seen as taking the theme park design collaboration that began in June and extending it across Disney. However, it cannot be confirmed that the 3D model and CAD ideas floated in June have all been put into practical use as of November. A distinction needs to be drawn between features already in use and those still under development.
At StudioLAB, the teams are prototyping and testing internal production processes. At ESPN Edge, they are exploring new ways to deepen relationships with sports fans and to make use of the archive of past footage. For both divisions, however, what has been announced is that joint projects are under way. It does not mean full automation of final film production or sports broadcasting, nor that a consumer-facing service is finished.
To support development in each division, Adobe will assign dedicated AI researchers and engineers. Disney will also get early access to new features in Premiere and Photoshop, in addition to Firefly and Foundry. In other words, the deal bundles not just AI models but also the specialists who support their development and the software used in everyday production. The contract value, any exclusivity terms, and the specific contract period have not been disclosed.
How Are Disney-Specific AI Models Built to Reproduce Its Worlds?
Firefly Custom Models and Firefly Foundry, both offered by Adobe, tailor AI models to a company's own assets, but they differ in scope. Firefly Custom Models is a feature that further trains a model on existing images to capture a particular style or subject. Firefly Foundry, by contrast, is a service in which Adobe researchers and engineers work with a company to develop a proprietary AI model, and it can extend to multiple concepts and media formats. In the Disney partnership, Foundry will be used to develop models that match character-specific traits, the backgrounds of individual works, and their art styles.
Firefly Foundry itself is a service announced on October 28, 2025, not a single new AI model developed for this partnership. Its distinguishing feature is that it supports the whole process, from developing an enterprise model to deploying it in real work. Adobe specialists co-develop the model with the company, evaluate its outputs, and set up controls over each user's access rights. It also integrates with GenStudio and Creative Cloud, and offers routes into production from Firefly and Adobe Express.
In the Disney development, artists lead the tuning of the model's expression. The aim is not simply to generate "Disney-style images" but to produce output that follows the traits of individual characters and the differing worlds, color palettes, and art styles of each work. The production side reviews generated images and can feed them into a process of revision as needed. Still, being able to further train a model on a specific style or character is not the same as being able to reproduce it accurately every time, and whether it is usable in production has to be judged along with an evaluation of its output.
The Foundry service as a whole covers not only images and video but also audio, vector graphics, and 3D. But what can be confirmed as actually in use at Disney is, for now, two image generation use cases at Imagineering. It does not mean that every media format Foundry supports has been deployed in Disney's production work.
How Does This Differ from Enterprise AI from FLUX and Google?
Adobe is not the only provider of mechanisms for further training AI models on a company's own assets. Black Forest Labs also offers fine-tuning and co-development using brand-specific assets with its enterprise FLUX, along with a choice of dedicated runtime environments. What deserves attention in the Disney partnership, then, is not the ability to build a dedicated model in itself, but how far the model can be tuned and how it is built into the production workflow. Based on each company's public information, here is a summary of what each offers and how it relates to the Disney deal.
| Option | Scope of model tuning | Deployment and operation | Relationship to Disney |
|---|---|---|---|
| Firefly Custom Models | Further training on existing images for a specific style or subject | Uses Adobe's custom model feature | Compared as a product category. Unclear whether it matches the implementation of Disney's dedicated models |
| Firefly Foundry | Co-development of a proprietary model with Adobe. Supports multiple concepts and media formats | Expert development support, evaluation, access management, integration with Adobe's creative apps | The framework for the joint development with Disney. Adoption across all media formats is unconfirmed |
| Enterprise FLUX | Fine-tuning and co-development using brand assets | Dedicated endpoints; eligible open-weight models can also run on a company's own cloud or on-premises | Unclear whether Disney compared or considered it |
| Plan to tune Google models via Foundry | Tuning Google models on Vertex AI with a company's own data | Adobe and Google announced the plan in October 2025 | Full implementation status and any adoption by Disney unconfirmed |
This is a comparison based on information each company had made public as of November 9, 2026. It does not evaluate image quality, generation speed, cost, or other factors under identical conditions.
The relationship between Google and Adobe is not a simple rivalry. On October 28, 2025, the two companies announced plans to integrate Gemini, Veo, and Imagen into Adobe's apps, and to let companies tune Google models on Vertex AI with their own data through Firefly Foundry. Google's models are thus both an alternative to Adobe and a technology that could be built into Adobe's creative environment. There has been no announcement, however, that Disney's dedicated models use a Google foundation model.
One feature of Black Forest Labs' FLUX is that eligible open-weight models can be run in a company's own cloud environment or on-premises. Adobe, for its part, includes everything from co-development with specialists to evaluation, access management, and integration with creative apps in the Foundry service. This difference leads to a choice about how much of model operation a company handles itself and how much it leaves to a development partner. That does not mean FLUX lacks ways to connect with Adobe products or rights-protection mechanisms, and no information has been published to suggest that Disney compared the two and chose Adobe.
The Training Data and Contracts Behind "Commercially Safe AI"
Adobe explains in its policy on training the Firefly foundation models that it uses licensed material, including Adobe Stock, and material whose copyright protection has expired. By contrast, for Disney's dedicated models, tuning uses material from characters and works owned by Disney itself, the rights holder. How a general-purpose foundation model was trained and what material is used to adapt a model to Disney's worlds need to be considered separately.
Adobe's "commercially safe" stance combines model design that takes into account brand rights and expressive requirements with contractual protections. However, a complete list of training material has not been made public, and there is no independent guarantee that rights infringement will not occur in any generated content. When companies actually use these tools, they need to check not only the model's training policy but also the terms of use and the scope of any indemnification.
Adobe's indemnification for intellectual property infringement for Firefly also comes with conditions on the eligible contracts, products, and features used. Indemnification is limited to customers who meet the stated conditions and whose contracts incorporate the relevant product description documents. What indemnification terms are set in this individual contract with Disney has not been disclosed, and it cannot be confirmed that the same indemnification applies to third-party models used through Adobe. For companies deploying dedicated models in their operations, how each feature is protected under the contract is an important factor in the decision, alongside which features are available.
Another effort to use Disney's intellectual property with generative AI is the agreement with OpenAI announced on December 11, 2025. It is a licensing arrangement that would let fans generate videos and images featuring Disney characters through Sora and other tools, so its target differs from that of the Adobe partnership, which supports internal production work. However, because the Sora web and app versions ended on April 26, 2026, and the API on September 24, it cannot be compared as a competing service operating as of November.
Months of Concept Development Cut to Hours; the Real Test Is the Downstream Impact
What stands out in this collaboration is the streamlining of concept development in theme park design. In an interview with Reuters, Disney's Kyle Laughlin explained that concept development that used to take months could be shortened to hours or days. The aim is to use AI to quickly create and compare multiple design options, and to find design problems and improvements before construction begins, reducing costly changes later in the process.
In theme park development, being able to compare many options early makes it easier to revisit a design before actual construction or equipment fabrication starts. However, the time savings Laughlin described are Disney's own account, and neither the number of projects covered nor the measurement conditions has been disclosed. And shorter concept development time does not mean the overall construction schedule shrinks by the same proportion, nor is it clear how much cost has actually been saved. Whether efficiency at the concept stage leads to less rework downstream will require further verification.
In judging the results, the share of AI-generated design options that were actually adopted for production will be important. It is also worth checking how much revision artists and designers had to make before adoption, and whether the results met the expressive requirements and world-building of the works. In addition, it will need to be assessed whether more trial and error at the concept stage reduced design changes in downstream processes such as construction and equipment fabrication.
The significance of this partnership lies not only in developing AI models that reproduce Disney's distinctive expression, but also in trying to build those models into the daily work of artists and designers and raise the efficiency of production as a whole. Being able to give shape to ideas quickly with generative AI is not the same as reducing actual production costs. If concrete results are shown in the future, such as the adoption rate of generated options, the time spent on revisions, and the number of changes in later stages, it will become clearer how much value enterprise-specific AI models bring to large-scale production environments.