Genativity

news

Disney picks Adobe to build Firefly Foundry models on its own IP, with no published training details

The multi-year alliance announced at Adobe MAX names Imagineering, StudioLAB and ESPN Edge as users, and discloses no contract value, no dataset size and no measure of how accurately the models reproduce Disney characters.

By Daniel Ochoa ·

Adobe and Disney announced a multi-year alliance on November 9, 2026 at Adobe MAX, naming Adobe as Disney's Official AI Innovation Partner. Adobe will use Firefly Foundry to build private models trained on Disney characters, environments and visual styles for Imagineering, StudioLAB and ESPN Edge. Financial terms, model ownership and any accuracy measurement were not disclosed.

Disney is putting its characters into private generative models built by Adobe, and the announcement setting that up carries no figure describing how closely the output matches the source material, no contract value, and no statement of who ends up owning the trained models.

The multi-year alliance was announced on November 9 at Adobe MAX in Miami Beach, with Adobe named Disney's Official AI Innovation Partner. Under it, Disney's creative teams get early access to Firefly, Firefly Foundry, Premiere and Photoshop, plus dedicated Adobe engineering and research staff assigned to Disney's production pipeline. Financial terms were not shared.

Firefly Foundry is the part that matters, and it is not something anyone else can run

Firefly Foundry is Adobe's managed service for building proprietary models for a single client, layering that client's catalog on top of Firefly base models. The Disney versions are customized on Disney properties and trained on character attributes, environments and visual styles, with output both companies describe as "commercially safe and production-ready." Adobe has not published the training method, dataset size or architecture of the Disney models, and Foundry is not a consumer product, so no outside party will be able to run the same prompt through it for comparison.

Imagineering has been using custom Firefly models since June

The theme-park side of this is already live. Walt Disney Imagineering Research and Development uses Firefly Foundry in its development pipeline, with a sketch-to-image model that converts hand-drawn concepts into rendered 2D concept art and a custom image model that produces franchise assets for properties including Mickey & Friends, Frozen, Moana, Lilo & Stitch and Cars. Disney described the training approach in June 2026 as a base model built on licensed and proprietary assets, then customized with more than 70 years of Imagineering concept art and technical designs. A 3D capability, turning 2D renderings into prototypes for build planning and materials estimates, was listed then as still forthcoming.

The new agreement extends that beyond the parks. Adobe becomes a StudioLAB innovation partner at Walt Disney Studios, where the first projects are described as custom models for post-production and VFX, and ESPN Edge will look at highlights, documentaries and uses for the ESPN archive. No film, no show and no specific effects work has been named.

"Franchise-accurate" is a quality claim with nothing behind it yet

Adobe and Disney both frame the models as "human-centered and artist-led," which is a statement about process rather than output. The claims that would actually be testable, that a model generates on-brand, franchise-accurate assets and production-ready results, arrive without a rubric, a human preference study, or a single score. For a model trained on one company's vault, accuracy to that vault is the only metric that counts, and it is exactly the metric nobody has published.

That gap is easier to see next to the models Firefly already hosts from outside partners. Google, Runway and Black Forest Labs systems sit on public leaderboards and can be compared prompt by prompt. In our benchmark, the same models can be scored on style adherence directly:

Style tests, Genativity benchmark
Google Veo 3.15.00 ±0.00
Runway Gen-4.54.93 ±0.21
FLUX 34.92 ±0.26
Genativity benchmark, 0 to 5, same prompts for every model, with 95% confidence interval. Full leaderboard

Those numbers exist because the models are generally available. A private Foundry model trained on Disney IP will not appear on any public board, which makes the internal evaluation Disney runs, assuming one exists, the only check on whether the output is usable. Two questions the announcement leaves open are worth tracking: who owns the custom models and their outputs, and whether anything Adobe learns here can travel into deals with other rights holders.

Sources: The Next Web, The Decoder, PPC Land, GamesBeat.

Cite this

Free to cite and reuse with a link back. Data is updated as new runs and prices come in, so include the date.

Genativity. (2026). Disney picks Adobe to build Firefly Foundry models on its own IP, with no published training details. Retrieved November 10, 2026, from https://genativity.com/news/disney-adobe-firefly-foundry-no-published-training-details
<a href="https://genativity.com/news/disney-adobe-firefly-foundry-no-published-training-details">Disney picks Adobe to build Firefly Foundry models on its own IP, with no published training details</a> (Genativity)

Daniel Ochoa

Daniel Ochoa: Covers model launches, shutdowns and pricing changes as they happen. Reads deprecation notices for a living so you do not have to.