Our Honest Review of Adobe Firefly

• Detailed images, videos, documents, and art from simple descriptions. 
• Text-to-edit capabilities that let you modify a photo or video's design via prompts.

• Paired with all the other design tools needed in the design workflow, like brushes, vectors, textures, etc. 
• Plans to integrate Firefly into its Creative Cloud

• Trained only on Adobe Stock data and openly licensed work 
• Non-commercial use only for the beta

In beta

The big friction point with AI image generators is copyright. You'll know if you care about this. If you don't. then you should probably be using Midjourney (it's more accessible, and it creates better outputs).

But if you do, you'll want to see what Adobe is doing with Firefly, its suite of generative AI in image and video.

Capabilities and features

Adobe rightfully understands that it can offer a lot when it comes to AI in images. So while "text to image" is a starting point, they're also working on:

• Recolor vector art
• Text to vector
• Text to pattern
• Sketch to image
• Text to brush

...and more.

Update 5/24/2023: Adobe has released Generative Fill, which is now available through the Adobe Firefly web app as well as Photoshop. It is very, very good! It matches its generations to the style and context of the image quite well. Hanne Van Briel did some very informative experiments on Twitter.

Copyright issues begone

The jury's still out when it comes to copyright responsibility in AI image generations.

And while that's the case, if you care about this particular issue (either out of principle or potential risk purposes) it's often safer to use Adobe's tools, which were trained on public domain and openly licensed images, along with stock images Adobe has access to via Adobe Stock.

Adobe is also working on tools to denote images that were created with AI, as well as ways to compensate users who contributed images that resulted in certain generations.

The quality trade-off

While the focus on licensing and transparency is admirable, note that Firefly's output is currently less impressive than Midjourney. That's just what happens when you have less training data at the start.

But it can get better. AI image generators often generate multiple takes at an image and log which one you choose (or none of them). They then use that data to get better.