Adobe Firefly Custom Models: Train an Image Generator on Your Own Style

Adobe has launched a public beta of Firefly Custom Models, a tool that lets brands and creators train image-generation models on their own material. The idea is simple: you upload your work (illustrations, photographs, characters, signature visuals) and the model learns to reproduce exactly that style and aesthetic. The result is a generator tuned to a specific visual language, rather than a general-purpose model you have to steer at length through prompts.

Adobe positions the tool primarily for teams that handle large volumes of content. Instead of starting from scratch every time, they get a reusable base: a model trained on approved internal assets. This matters most for brand compliance. Marketing teams, publishers and agencies with ongoing clients get a tool that keeps visual consistency across projects automatically.

Custom Models cover three main areas: characters, illustrations and photographs. That means the tool targets animation and game studios, which need precision in reproducing specific characters, as well as brands that want photorealistic generation in a recognizable look.

Unlike the general-purpose Firefly, Custom Models is a closed system when it comes to data. Adobe has repeatedly stressed that Firefly is trained only on licensed material from Adobe Stock and public-domain content. With custom models, the user supplies the data, which removes copyright questions about the output.

The move to public beta means the tool is available without waiting in a queue; all you need is the appropriate Creative Cloud subscription. Adobe promises to expand its capabilities as it works through feedback.

For the market, this is a signal: customizing generative models for brand needs is moving from "niche development" to a standard product tool. Midjourney, Stable Diffusion and others already offer fine-tuning options, but Adobe is betting on simplicity and legal cleanliness, which is exactly what corporate customers lack.