Disney commissions custom Firefly models from Adobe, trained on its characters

If you own a brand with recognizable characters, you know the biggest headache with generative AI: a general-purpose model draws something that looks "roughly right," and lawyers then spend weeks figuring out whose rights have been infringed. Disney may be the world's intellectual property owner most sensitive to this problem. Now the company has chosen its approach: commission its own model from Adobe instead of using someone else's.

What was announced

The companies unveiled a multiyear partnership at Adobe MAX. Disney gets early access to new Adobe technologies for its theme parks, cruise line and ESPN content. The key tool is Adobe Firefly Foundry. According to Adobe, it will be used to build custom models for Disney brands. These models will know the company's characters, environments and visual styles and deliver "commercially safe," production-ready output.

The deal spans several areas. Imagineering, Disney's research division, already uses Firefly in park development. Adobe will expand internal creative workflows at Walt Disney Studios. The agreement includes the studio's StudioLAB and ESPN Edge, ESPN's innovation hub. There, AI is intended to increase fan engagement and find new uses for ESPN's archive.

Why a brand needs its own model

Imagine an artist developing a concept for a new attraction. With a standard generator, they have to describe the character from scratch every time, argue with the model over proportions and costumes, then manually remove anything reminiscent of someone else's style. A model trained on one company's material should address three problems at once: visual consistency, less manual editing and a clearer legal status for the output.

This is our interpretation, not a quote from the announcement. But that is the logic behind Adobe's approach to "safety": the bet is that the risky part of the work, the rights to the training data, is covered by a contract rather than hopes of favorable court rulings.

The choice of supplier is worth noting, too. Disney previously struck a deal with OpenAI but walked away from it. The company also figured in the story surrounding Character.AI's bankruptcy. Against that backdrop, Adobe looks like a conservative choice: a major vendor, familiar tools in creative pipelines and an emphasis on predictability rather than the flashiest demo.

What we don't know

The original announcement provides few technical details. We don't know how training works: whether it involves fine-tuning Firefly's foundation models or something more extensive. It is unclear where Disney's data is physically stored and whether it feeds into Adobe's general-purpose models. There are no quality metrics, timelines or prices. For now, this is a statement of intent and a partnership, rather than a product that can be tested.

There is also some uncomfortable context. According to Bloomberg, Disney plans to cut thousands of jobs and reduce costs by 20% annually. This would be the company's fourth round of layoffs in a year. The stated goal of AI is to strengthen creative workflows, but it is hard to miss that investments in automation are going hand in hand with savings on staff. Disney has not publicly explained exactly how the two processes are connected.

What practitioners should take away

Product managers and founders should take three things from this story.

• A custom brand model is becoming a standard enterprise product. If you build tools for media, marketing or e-commerce, demand for "a model that knows our style" will grow, and general-purpose generators are at a disadvantage here.

• Legal clarity is being sold as a feature. The phrase "commercially safe" has become part of the positioning. If you build on other companies' models, think ahead about how you will answer clients' questions about the origins of the training data.

• Major players choose predictability. Disney switched partners, and its choice of Adobe suggests that integration into familiar pipelines and contractual arrangements matter more in the enterprise than benchmark leadership.

The next thing to watch is whether Firefly Foundry produces public case studies and terms for smaller companies. If Adobe turns these models into a clearly defined business product, the market for custom image and video generation will become considerably more crowded. For now, this is a strong signal of where things are heading, but no proof that the approach works at scale.