Hands-on with Adobe Firefly, their entry into the crowded AI-generated image space

    Adobe’s creative suite of applications could be among the most disrupted by the influx of new AI-powered applications.

    Rather than sit back and wait, Adobe has entered the game with a service (currently in beta) called Firefly.

    For those who’ve used MidJourney to create AI-generated artwork, this will feel familiar, but I actually really appreciate the speed and options available with Firefly.

    To start, you get to select the aspect ratio you need, with choices between landscape (4:3), Portrait (3:4), Square (1:1) and Widescreen (16:9). This enables you to create art for a range of purposes ranging between social media posts, blog posts, desktop backgrounds and more.

    To use the service, you simply chose between the two available experiments – Text to image and Text effects, then input a search term and wait a few seconds for the original image to be created.

    It seems Adobe has a big roadmap for Firefly, with other features like Recolor Vectors coming soon and plenty more in development. Features being explored include:

    • Inpainting – Use a brush to add, remove, or replace objects in an image. Generate the new fill with a text prompt.
    • Personalized results – Generate images based on your own object or style.
    • Text to vector – Generate editable vectors from a detailed text description.
    • Extend image – Change the aspect ratio of your image with a single click.
    • 3D to image – Generate images from the interactive positioning of 3D elements.
    • Text to pattern – Generate seamlessly tiling patterns from a detailed text description.
    • Text to brush – Generate brushes for Photoshop and Fresco from a detailed text description.
    • Sketch to image – Turn simple drawings into full-colour images.
    • Text to template – Generate editable templates from a detailed text description.

    Hands-on with Adobe Firefly (beta)

    To test the service, I came up with a few search ideas myself but also thought it was a good opportunity to visualise the creative thoughts of my 5yo daughter. Here is a couple of ideas she came up with.

    Family with sun and clouds and rainbow and unicorn and flowers and a beautiful sunset

    The results are as below. What’s great about the results is that they met the criteria of the text-to-image search, with plenty of variety to choose from. This combination of real and fictitious elements was actually a great test of the AI, to see how well it blended these elements together to make a semi-believable image.

    What is apparent is that where humans are being represented, there are glitches in their faces and hands.

    An owl sleeping in the tree

    With this request being one based on realism, the AI-generated 4x photo-realistic images. Again there’s a great variety of options to choose from, with the owl in different positions, and even different framing in the shots.

    If I wanted a different feel to the images, I could take my shot at hitting the ‘refresh’ button to roll the dice, or if I want more control, then select a style from the panel on the right. These styles contain a wide variety of themes including Steampunk, Cyberpunk, Psychedelic, Product photo, Hyper-realistic, Cartoon and many more.

    Drone flying inside an empty warehouse in front of clouds of coloured smoke

    Now it was my turn. What impressed me about the images created for this query is that the lighting on the drone makes it look like the drone is really in an indoor environment. The coloured smoke effect looks amazing and without being asked, the AI has made segments of the smoke coloured, but a portion of it is normal. This gives it a realistic feeling, with a great look.

    The images are made even better with a nice depth-of-field effect.

    What isn’t great, is the actual detail on the drone. While all the main components of the quad are there, the props, the camera sensors, and the cameras, however, it looks like the AI was trained hard on DJI drones as the characteristics are incredibly close to DJI drones, but right before the end of the render, it seems the AI realised and erased part of the drone to make something unique.

    Unique house design with modern minimal architecture that would win Grand Designs

    Home designs are another example of a text-to-image request that results in photo-realistic results. Entering a request like this, requires the AI to break the request into a number of sections. Firstly the request for a house design, then a unique house design, then requires it to understand what modern, minimalist design is, and finally understanding what the Grand Design TV show is and what a standout ‘winning’ design would be.

    The result is seriously impressive and while a Google search may yield similar modern home designs, these do appear unique and we got not only just a house, but completed homes, with amazing landscaping.

    A butterfly on a beer bottle

    While this request technically could happen, it’d be incredibly rare and I imagine the reason we got such a colourful and whimsical response to this, is a determination that Firefly needed to create a creative set of images, rather than a strictly photo realistic one.

    The variety of colours and different zoom levels are really appreciated, offering a good variety of images to choose from.

    A rockband from the year 3000 cyberpunk style

    My final request was one that again had multiple components that needed to be first understood, then created and blended together.

    My favourite out of this four is the lower right. When you’re happy with an image, you can select it and download a larger image. These images are not designed for commercial use, therefore are watermarked in the lower-left corner when downloaded.

    The request for a band could have resulted in an image containing band members in normal attire set against a cyberpunk background, but instead, the AI did a great job of creating a Cyberpunk band in a neon-filled environment complete with futuristic glasses.

    Jason Cartwright
    Jason Cartwright
    Creator of techAU, Jason has spent the dozen+ years covering technology in Australia and around the world. Bringing a background in multimedia and passion for technology to the job, Cartwright delivers detailed product reviews, event coverage and industry news on a daily basis. Disclaimer: Tesla Shareholder from 20/01/2021

    Leave a Reply


    Latest posts


    Related articles