Adobe opens Firefly Custom Models public beta for on-brand images

In short: Adobe has started a public beta of Firefly Custom Models, which lets creators and brands train Adobe’s image AI on their own pictures to keep a consistent look.

What happened

Adobe’s Firefly Custom Models let you upload your own images and use them to teach Firefly what your brand, characters, or art style should look like. You can then type a text prompt and ask Firefly to generate new images that match that same style, like giving the AI a small “style guide” it can follow.

Adobe says the training process uses a small set of high-quality JPG or PNG files. The images need to be at least 1024 by 1024 pixels, under 50 MB, and in common wide or tall shapes. After training, teams can preview results, test prompts, refine the model, and share it so everyone is making content that looks similar.

The company is also tying Custom Models into its other tools, including the Firefly web app and Adobe GenStudio, which is used to make marketing content like ads, emails, and social posts. Over time, Adobe plans broader support beyond images, including video, audio, vector graphics (shapes that scale cleanly like logos), and 3D.

Why it matters

Many people using image generators end up with results that look inconsistent from one picture to the next, especially for recurring characters or branded campaigns. Custom Models are meant to reduce that problem, which can save time and money for teams that need lots of visuals that all look like they came from the same playbook.

Source: The Verge AI