Adobe Launches Firefly Custom Models in Public Beta

Adobe is launching customizable AI image generators that can mimic specific artistic styles and character designs. The Firefly Custom Models are available in public beta as of March 19, 2025, allowing creators and brands to train a model on their own assets to ensure generated images follow a consistent aesthetic.

What the tool does

Creators can feed their own images into custom models that learn to mimic specific styles and aesthetics. Adobe says the tool preserves details like stroke weight, color palettes, lighting, and character features across generations. Custom models are private by default, meaning images used to train them won't be fed into Adobe's general Firefly models.

The feature aims to streamline workflows for teams producing high volumes of content. Instead of starting from scratch with each new project, creators can reuse a trained model across multiple briefs and campaigns while maintaining visual consistency.

Background and ethical positioning

Adobe first announced Firefly Custom Models as a private beta at Adobe Max 2024. Adobe has consistently promoted its Firefly models—trained using licensed and public domain content—as an ethical alternative to competitors that may have scraped protected works.

This expansion gives creative professionals more granular control over their AI training data. However, Adobe's current safeguards rely on user attestation. According to Adobe's help documentation, users must confirm they have the necessary rights and permissions before training a custom model and that "your use of custom models won't infringe on the copyright, IP, likeness, or privacy rights of others." Adobe has not disclosed what enforcement mechanisms prevent users from training models on copyrighted work they don't own.

What this means

Firefly Custom Models address a real workflow problem for studios and brands: maintaining visual consistency at scale. The private training approach is a meaningful differentiator from general-purpose models. However, the reliance on user attestation rather than automated enforcement leaves open the possibility of misuse. This feature will likely appeal to small studios and brand teams, but Adobe's enforcement burden—or lack thereof—remains an open question for creators concerned about unauthorized model training.