Starting today, Adobe's Firefly image generator supports custom model training. The new Firefly Custom Models feature, now in public beta, lets creators and brands train the AI on their own images so the output matches their specific visual style.
This fills a real gap. Generic AI image generators produce generic-looking output. A brand with a distinct visual identity - specific color palettes, illustration styles, character designs - gets results that look like everyone else's. Custom training changes that equation: feed in your existing asset library, and generated images actually look like they belong in the same campaign.
Two Use Cases, One Goal
Adobe is targeting two scenarios. First, style consistency: train on your brand's visual language so every generated image follows the same aesthetic across campaigns and channels. Second, character consistency: train on specific character designs so they look the same in every generated image. That matters for anyone doing serialized content, mascot-driven branding, or product illustrations.
Both solve the same underlying problem. AI image tools are fast at generating volume, but terrible at staying visually coherent without heavy manual correction. Custom training aims to fix that at the model level rather than the editing level.
Late to the Party, but Better Connected
Midjourney has offered style references for months. Stable Diffusion users have been fine-tuning (training a model on a small dataset to specialize its output) with LoRA adapters since 2023. Adobe is not first here.
But Adobe's advantage isn't the feature itself - it's where the feature lives. Custom Models plugs directly into Creative Cloud, which means outputs flow straight into Photoshop, Illustrator, and InDesign without leaving the Adobe environment. For creative teams already paying for Creative Cloud, this removes the friction of using a separate AI tool and importing results back into their actual production pipeline.
The public beta is available now to Creative Cloud subscribers. Adobe hasn't announced whether the final version will require an additional subscription tier or ship as part of existing plans.