Adobe Firefly

Updated for version: Firefly Image 3

Accessible via: https://firefly.adobe.com/

Requires login via: Google, Apple, Facebook, Microsoft, Line

Ratings

Accuracy / Quality ★★☆☆☆

Flexibility / Features ★★★☆☆

Data security / Privacy ★★★★☆

Pros/cons

Pros 

  • Trained on data within Adobe’s own copyright.
  • Integration into other Adobe products possible.
  • Explicit examples/controls for modification of images such as style, angle and lighting.

Cons 

  • Limited number of image generations per month.
  • Less capable than comparable models with text in images, faces, or limbs.
  • More limited understanding of prompts.

Description

Adobe Firefly sets itself apart as image generator through the training data used. Most commercial AI models are trained on a large volume of images that are available online, regardless of whether the company has the right to use these images or not. This aspect has led to several court cases already, with the right to use images for the training of AI being challenged. To avoid this problem Adobe decided to take the more legally safe route and train their image model using only images that are in the public domain (where the copyright has expired) and images to which they own the copyright themselves. As a result the images generated by Adobe Firefly are more safely usable, but the quality of the images is lower due to a smaller volume of training data.

Features and examples

Composition, styles and effects

When an image is created in Adobe Firefly (Text to image mode) you are presented with the option to modify the image. These modifications are explicit and also include a visual representation of what that modification would result in. These include:

1. Content type

2. Composition

3. Styles

4. Effects

Generative expand / Generative fill

These features are currently in preview only and are not fully available to the public yet.

Beyond the Text to image mode of Firefly, there is also the option to ‘expand’ images or to replace objects in images with the ‘fill’ option. With Generative Fill you upload an image, select the area of the image you want to replace, and Firefly is then used to re-create this selected area using your prompt as guidance. With Generative Expand an image can be increased in size by generating ‘more’ of that image on the side. This is done without a textual prompt, meaning that Firefly uses the existing image as guidance to generate the remainder of the image space.