> ## Content Index
> Fetch the complete content index at: https://www.cognativ.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Adobe Firefly Foundry: Adobe AI Foundry Builds Custom Generative AI Models for Enterprises
- URL: https://www.cognativ.com/blog/adobe-firefly-foundry/
- Published: 2026-10-31T07:57:10.000Z
- Updated: 2026-11-08T11:15:04.000Z
- Description: Adobe launched AI Foundry in October 2025 to deep-tune Firefly on brand IP. See who uses it, how it handles ownership, and when lighter custom models fit.
- Author: Kevin Anderson
- Tags: News

Adobe Firefly Foundry is Adobe's service for building private generative AI models that are trained on a company's own intellectual property. Adobe first introduced it on October 20, 2025 as Adobe AI Foundry, then formally announced it as Adobe Firefly Foundry at Adobe MAX on October 28, 2025\. Instead of handing businesses a self-serve tool, Adobe pairs them with its own scientists and engineers, who deeply tune Firefly models on a brand's catalog of images, video, audio, vector art and 3D assets.

This guide covers what Firefly Foundry is, how the launch unfolded, how the service works, how it compares with the lighter Firefly custom models, and what enterprise teams should weigh before committing. Facts are refreshed through October 31, 2026, and later developments are dated so you can see what changed after launch.

**Key takeaways**

- Adobe AI Foundry and Adobe Firefly Foundry are the same offering; the Firefly Foundry name took over at Adobe MAX 2025.
- Foundry models are multimodal, multi-concept and built with Adobe experts; Firefly custom models are self-serve and image-only.
- Named users include The Home Depot, HUMAIN and Walt Disney Imagineering, and Adobe has since pushed Foundry into film and TV production.

## What Is Adobe Firefly Foundry?

In its [October 28, 2025 announcement](https://news.adobe.com/news/2025/10/adobe-max-2025-firefly-foundry?ref=cognativ.com), Adobe describes Firefly Foundry as a way for businesses to "work directly with Adobe" to create generative AI models unique to their brand. The models are trained on entire catalogs of existing IP and built on top of Adobe's commercially safe Firefly models. Adobe says they can support all major asset types: image, video, audio, vector and 3D.

The offering has three parts. First, the Foundry models themselves, which Adobe trains with each business on its existing IP. Second, an application to manage and deploy those models, including testing outputs and controlling who in the organization can use them. Third, co-innovation: embedded Adobe experts, including applied AI and machine learning scientists and forward-deployed engineers, who help design and ship specific use cases.

Once trained, a Foundry model is surfaced inside the tools creative and marketing teams already use. Adobe lists GenStudio, Creative Cloud, the Firefly app and Express as integration points. That matters because a custom model is only useful if designers and marketers can reach it without leaving their normal workflow.

![A private Firefly Foundry model at the center connects to image, video, audio, vector and 3D outputs.](https://storage.ghost.io/c/cc/dc/ccdca949-0fe4-4119-8990-70a44ad75060/content/images/2026/10/adobe-firefly-foundry-1.png)

One deeply tuned Foundry model can produce on-brand image, video, audio, vector and 3D assets.

## From Adobe AI Foundry to Firefly Foundry: The Launch Timeline

The two names cause most of the search confusion around this product. [TechCrunch reported on October 20, 2025](https://techcrunch.com/2025/10/20/adobe-launches-a-foundry-service-that-builds-custom-generative-ai-models-for-enterprises/?ref=cognativ.com) that Adobe had launched "AI Foundry," a service that builds custom generative models on a company's branding and IP. Eight days later, Adobe's official MAX release used the name Adobe Firefly Foundry, and that is the name Adobe has used since.

### Launch week, October 2025

At launch, Adobe told TechCrunch that enterprises had created more than 25 billion assets with Firefly models since Firefly debuted in 2023, and that Foundry would be priced on usage rather than per seat. Hannah Elsakr, vice president of GenAI New Business Ventures at Adobe, framed it as a response to demand: enterprises had asked Adobe to advise them and partner with them, not just license software. The MAX release also said the team from Invoke, a generative media company, had joined the Firefly Foundry team.

### What has happened since

On January 22, 2026, Adobe published [new media and entertainment partnerships for Firefly Foundry](https://blog.adobe.com/en/publish/2026/01/22/adobe-partners-artists-power-new-era-media-entertainment-firefly-foundry?ref=cognativ.com), naming talent agencies CAA, UTA and WME, studios such as B5 Studios and Promise Advanced Imagination, and directors David Ayer and Jaume Collet-Serra. The same post said brands including The Home Depot, HUMAIN and Walt Disney Imagineering were already working with Firefly Foundry.

On June 16, 2026, Adobe announced a [collaboration with Walt Disney Imagineering](https://news.adobe.com/news/2026/06/adobe-and-disney-imagineering-collaborate?ref=cognativ.com). According to Adobe, Imagineers will use Foundry models built on Imagineering's own design catalog for sketch-to-image concept art, franchise-accurate image generation and turning 2D concepts into 3D prototypes. As of October 31, 2026, Adobe had not published public list pricing for Firefly Foundry.

![Timeline: AI Foundry launch, renamed Firefly Foundry at Adobe MAX, film and TV partners join, Disney Imagineering collaboration.](https://storage.ghost.io/c/cc/dc/ccdca949-0fe4-4119-8990-70a44ad75060/content/images/2026/10/adobe-firefly-foundry-2.png)

Adobe AI Foundry became Firefly Foundry at MAX, then expanded into film, TV and theme park design.

## How Firefly Foundry Works

The workflow starts from Adobe's Firefly models, which Adobe says were trained on licensed data. Adobe then retrains them on the customer's proprietary content and brand guidelines, packages the result as a private model and deploys it into Adobe applications.

### Deep tuning rather than fine-tuning

Adobe draws a deliberate line between its approach and ordinary fine-tuning. In an [October 2025 interview with VentureBeat](https://venturebeat.com/ai/adobe-foundry-wants-to-rebuild-firefly-for-your-brand-not-just-tweak-it?ref=cognativ.com), Elsakr said Adobe is "literally retraining our base model," which is why it calls the process deep tuning. Lightweight methods such as adapters change a small set of weights on top of a frozen model; our explainer on [how LoRA fine-tuning works](https://www.cognativ.com/blog/what-is-lora-llm/) covers that trade-off. Deep tuning aims for broader brand understanding at the cost of more compute and expert time, which explains why Foundry is a service rather than a button.

### Multimodal brand models

Because Foundry models can learn many concepts at once, a single model can carry a brand's characters, products, color systems and visual style across formats. A retailer could produce product imagery, short video and localized campaign variants from one model, while a studio could keep a franchise's look consistent from storyboards through post-production.

### Deployment and access control

The management application lets teams test outputs before release and decide which groups can use the model. For most enterprises the harder work sits around that application: connecting asset libraries, approval steps and ad delivery systems so generated content moves through the same checks as human-made content. That is where an [AI integration partner](https://www.cognativ.com/ai/integration/) typically helps, linking the model to existing DAM, CMS and marketing platforms.

![Brand IP catalog and the Firefly base model feed deep tuning by Adobe experts, producing a private Foundry model deployed into Adobe apps.](https://storage.ghost.io/c/cc/dc/ccdca949-0fe4-4119-8990-70a44ad75060/content/images/2026/10/adobe-firefly-foundry-3.png)

Adobe deep tunes Firefly on a company's IP, then deploys the private model into the apps teams already use.

## Firefly Foundry vs Firefly Custom Models

Adobe offers two ways to customize Firefly, and the "firefly foundry vs firefly custom models" question comes down to scope and who does the work. Firefly custom models were announced in private beta at MAX 2025 and, according to [Adobe's March 19, 2026 post](https://blog.adobe.com/en/publish/2026/03/19/adobe-firefly-expands-video-image-creation-with-new-ai-capabilities-custom-models?ref=cognativ.com), moved to public beta. They let a user train a model on their own images to capture a style, character or photographic look, and Adobe says they are private by default.

| Aspect          | Firefly Foundry                    | Firefly custom models                     |
| --------------- | ---------------------------------- | ----------------------------------------- |
| Who builds it   | Adobe experts with the business    | The user, self-serve                      |
| Training method | Deep tuning of the base model      | Training on a small set of user images    |
| Concepts        | Multiple concepts in one model     | One style, subject or character           |
| Output formats  | Image, video, audio, vector and 3D | Images                                    |
| Typical buyer   | Large enterprises and studios      | Creative pros and teams                   |
| Pricing basis   | Usage-based, negotiated            | Self-serve within Firefly, in public beta |

In short, custom models suit a design team that needs consistent illustrations or a recurring character quickly. Foundry suits an organization whose content spans many formats and whose IP, such as a franchise or a product catalog, is valuable enough to justify a dedicated, expert-built model.

## IP Ownership and Commercial Safety

Commercial safety is Adobe's main differentiator. Firefly base models are trained on licensed content, and Adobe says Foundry adds the customer's own catalog on top. Elsakr told VentureBeat that Adobe keeps each enterprise's IP separate and never feeds it back into the base model, and that "the enterprise itself owns that output." Adobe's Disney release contrasts this with models trained on scraped internet data, which it says offer no guarantees on IP fidelity or provenance.

Those are vendor commitments, so buyers should confirm them contractually: data retention, model ownership on contract exit, indemnity terms and where training runs. The same questions apply to any model trained on sensitive material, as our guide to [using an LLM with private data](https://www.cognativ.com/blog/how-to-use-llm-with-private-data/) explains.

## Firefly Foundry Pricing and Access

Adobe has not published a price list. TechCrunch reported at launch that Foundry is offered on a usage basis rather than per seat, unlike much of Creative Cloud, and access runs through Adobe's enterprise sales and Foundry teams rather than a sign-up page. Expect a scoped engagement covering data preparation, model training, deployment and ongoing usage.

The early customer list hints at who Adobe is targeting: a large retailer, an AI company, a theme park design group and film and TV producers. These are organizations with deep visual archives and high content volumes, where a model that already knows the brand can save repeated review cycles.

## What Firefly Foundry Means for Enterprise AI Teams

Firefly Foundry reflects a wider shift: enterprises want models shaped by their own data instead of relying only on general-purpose systems. The same pattern appears in text, where companies weigh retrieval, fine-tuning or [a formal AI governance framework](https://www.cognativ.com/blog/enterprise-ai-governance-framework/) before scaling generative tools across departments.

A practical path for evaluating Foundry has four steps. Audit the brand assets you own and the rights attached to them. Define two or three use cases with measurable outcomes, such as campaign localization time. Set governance rules for approvals, disclosure and model access. Then pilot, compare results with your current process and decide whether the deeper Foundry engagement or lighter custom models deliver enough value.

Foundry also does not remove the need for broader architecture choices. Generated assets still need metadata, rights tracking and links into the systems that publish them, and teams that already run text models may want one approach to model selection across both. Our [LLM solutions practice](https://www.cognativ.com/ai/llm-solutions/) works on that layer for language models, and the same discipline applies to image and video models.

![Four steps: audit brand assets, define use cases, set governance rules, pilot and measure.](https://storage.ghost.io/c/cc/dc/ccdca949-0fe4-4119-8990-70a44ad75060/content/images/2026/10/adobe-firefly-foundry-4.png)

Evaluate Foundry by auditing your IP, defining use cases, setting governance and running a measured pilot.

## Frequently Asked Questions

### What is Firefly Foundry?

Firefly Foundry is an Adobe service that builds private generative AI models for businesses. Adobe experts deeply tune Firefly models on a company's own IP so the models can generate on-brand images, video, audio, vector art and 3D content inside Adobe tools.

### Is Adobe AI Foundry the same as Firefly Foundry?

Yes. Adobe introduced the service as AI Foundry on October 20, 2025, and formally announced it as Adobe Firefly Foundry at Adobe MAX on October 28, 2025\. Firefly Foundry is the current name.

### What is the difference between Firefly Foundry and Firefly custom models?

Custom models are self-serve, image-only and focused on one style, subject or character. Foundry models are built with Adobe experts, cover multiple concepts and support image, video, audio, vector and 3D outputs.

### How much does Adobe Firefly Foundry cost?

Adobe has not published list pricing. At launch, Adobe said Foundry is priced on usage rather than per seat, and engagements are arranged through Adobe's enterprise teams.

### Who owns content created with Firefly Foundry?

Adobe has said the enterprise owns the output and that customer IP is kept separate from the Firefly base model. Buyers should still confirm ownership, retention and indemnity terms in their contract.

### Which companies use Firefly Foundry?

Adobe has named The Home Depot, HUMAIN and Walt Disney Imagineering, along with media partners such as CAA, UTA, WME, B5 Studios and Promise Advanced Imagination.

## Conclusion: Is Adobe Firefly Foundry Right for Your Business?

Adobe Firefly Foundry turned the October 2025 AI Foundry launch into a lasting enterprise offering: private, deeply tuned, multimodal models built on commercially safe Firefly foundations and delivered with Adobe's own experts. A year on, its customer list has grown from retail to theme parks and film production, while Firefly custom models give smaller teams a lighter, self-serve option.

The right choice depends on how much IP you own, how many formats you produce and how ready your workflows and governance are for generated content. If you are weighing a custom generative model, or need it connected to the platforms your teams already use, [talk to Cognativ about your AI project](https://www.cognativ.com/contact/).