Custom Model

Custom models represent a class of AI systems designed to be trained or fine-tuned on domain-specific or user-provided data, allowing them to replicate particular styles, visual identities, or behavioral patterns that generic models cannot reliably reproduce. Rather than relying solely on broad pre-training across open-ended internet data, these models are built to internalize the nuances of a custom dataset—whether that consists of brand imagery, character art, photographic aesthetics, or specialized conversational tone. This design intent makes them especially valuable for workflows where consistency and stylistic fidelity matter more than raw generality, such as maintaining a recognizable look across large content campaigns or preserving character design integrity across illustration projects.

The training lineage of a custom model typically begins with an open or proprietary foundation model, which then undergoes supervised fine-tuning on curated user assets to cultivate style embeddings, visual markers, or task-specific behaviors. In the Adobe Firefly ecosystem, this process is made accessible through a public beta where creators upload their own images to produce reusable models that can be applied repeatedly within creative pipelines. Once trained, these models are deployed alongside other industry-leading models in multi-model creative studios, giving teams the ability to switch between or blend custom and general-purpose outputs depending on project needs. This combination of flexible fine-tuning and deployment within larger AI studio environments positions custom models as a practical bridge between one-off generative outputs and scalable, on-brand production.