Adobe drops Firefly Image 2 and new vector AI model

Generative AI images have always looked impressive from afar, but the moment you try to use them in professional production, the headaches begin: the styling drifts between prompts, and every output is a flat, unscalable bitmap that requires tedious vector tracing.

Adobe tackled both problems head-on at its annual MAX conference with a trio of new foundation models: Firefly Image 2, the Firefly Vector Model, and the Firefly Design Model. According to Adobe, users have already generated over 3 billion images since Firefly first entered beta in March.

Native vectors with actual editable paths

The marquee debut is the Firefly Vector Model, arriving as a Text to Vector Graphic beta inside Illustrator on desktop. Adobe calls it the first generative model built specifically for vector art, trained on tens of millions of vector graphics from Adobe Stock.

Unlike traditional auto-trace tools that drown you in thousands of messy anchor points, Adobe VP Alexandru Costin noted the team specifically optimized the model to generate paths with the fewest points possible. The resulting scenes come neatly grouped, layered, and feature editable gradients rather than a chaotic pile of shapes. Illustrator also added a Mockup tool to project vector art onto 3D objects, alongside a web version of the app.

Style cloning and manual camera controls

Over in the standalone web app, Firefly Image 2 brings a massive bump in photorealism, particularly across complex textures like skin pores and foliage. It now incorporates manual photo settings, letting creators tweak shutter effects, motion blur, and depth of field.

It also introduces Generative Match. Instead of writing endless adjectives trying to recreate a bespoke aesthetic, users can upload a brand reference image and use a slider to control how closely the output adheres to that visual identity. Meanwhile, Adobe Express received the Firefly Design Model for text-to-template creation, intelligently positioning copy and complementary color palettes across standard print and digital aspect ratios.

Why it matters

We think Adobe's real moat isn't model architecture; it is workflow integration. Competitors like Midjourney produce stunning isolated renders, but bringing those assets into production usually demands hours of cleanup and re-drawing. By rendering native, layered vectors directly on an Illustrator artboard, Adobe turns generative output into immediate utility for commercial graphic artists.

Generative Match solves another major enterprise gripe: visual consistency. Marketing teams can finally mass-produce campaign assets while strictly maintaining brand coherence, backed by Adobe's legal IP indemnification and default Content Credentials metadata.

There are still limitations. Generating clean vector geometry for intricate technical diagrams remains difficult, and complex prompt reasoning will take time to refine. But by embedding AI directly where professionals already do their billable work, Adobe has effectively raised the floor for commercial design speed.