Adobe’s Firefly 3 lets you skip the prompt in Photoshop

If you have ever spent an hour rewriting a paragraph of text just to get an AI image generator to match the exact shade and mood you want, Adobe feels your pain.

Adobe CTO of Digital Media Ely Greenfield put it bluntly: prompting is a pain in the butt. Instead of sweating over three paragraphs of adjectives, designers can now just show Photoshop what they mean.

Adobe unveiled its next-generation Firefly Image 3 foundation model and rolled out a batch of new generative AI tools inside the Photoshop desktop beta. The features are available today for beta testers, with general availability slated for later this year, alongside a public web beta for Firefly.

Show, don't tell

The headline feature is Reference Image. Instead of describing a complex visual style through text prompts alone, users can upload an existing picture to steer the AI output, matching its colors, texture, and aesthetic.

Naturally, using reference images raises legal questions. Adobe says users are expected to hold the rights to whatever they upload, and a warning will flag this ownership requirement upon first launch. Adobe noted that uploaded references will not be used to train Firefly, and the company is developing a universal "do not train" tag that can also block images from being referenced.

Even with the user shouldering responsibility, Adobe still claims the output remains safe for commercial use.

Photoshop beta also gained Generate Background, tailored for quick product photography swaps, Enhance Detail to sharpen soft generations, Generate Similar to iterate on favorite outputs, and Generate Image, which lets users generate a full canvas from scratch.

Meanwhile, the underlying Firefly Image 3 model brings sharper lighting, better spatial positioning, clearer in-image text rendering, and improved comprehension of long prompts. Standard tools got love too, including a non-destructive Adjustment Brush and instant access to over 25,000 cloud fonts.

Why it matters

We think Adobe is making a very smart concession here: pure text prompting is often an awful interface for visual professionals.

In real design workflows, nobody wants to play prompt roulette for an hour when they already have a client deck, mood board, or product photo. Giving users reference-based control makes generative AI practical for everyday studio work, especially for e-commerce sellers swapping backdrops.

However, this flexibility introduces potential friction. Adobe built Firefly’s entire brand on being clean and commercially safe compared to Midjourney or Stable Diffusion. Shifting the burden of copyright ownership to users for reference imagery opens the door to messy IP gray zones, even if Adobe insists uploaded files stay out of the training loop.

Still, by weaving reference tools, non-destructive adjustment brushes, and massive font libraries directly into the Photoshop canvas, Adobe is tightening its grip on professional workflows. Rivals might generate flashier standalone images, but Adobe is making sure you never actually need to leave Photoshop to get paid work done.