Midjourney and its alternatives: a guide to AI image generators

AI-based image generation has been evolving unbelievably fast in the last couple of years, making it increasingly harder to tell real and generated images apart. 
The generative AI platform Midjourney, developed by an independent San Francisco research lab, recently shut down free trials of their software after a fake picture of Pope Francis wearing a puffer jacket went viral, along with another set of images purportedly showing a true-to-life yet never-happened violent arrest of Donald Trump. On top of that, just a few days ago, the AFP had to put out a statement to confirm that another viral picture, showing an elderly French protester with his face covered in blood after an altercation with the police, was actually produced using MidJourney. The ability to fool the large public is a byproduct of the improved accuracy of the recently released version 5 of Midjourney, which fixed some of the most glaring mistakes the AI was still disseminating into the generated images. Human hands, for example, had always been a giveaway of the artificial nature of the pictures produced by the generative model. That’s no longer the case, and the consequences have quickly revealed themselves. 
While it’s becoming clear that the ability of these models will have lasting consequences on the role of media in society, the development of Midjourney and other competing AI generators seems unstoppable. Here are three alternatives to the service that have been growingly used to generate illustrations and pictures and promise to become better and better in the coming months.

While it’s known mostly for its large language model GPT and the chatGPT web interface, OpenAI has been working on one of the most widely used image generators, Dall•E 2. Like Midjourney, OpenAi’s software can create images in a seemingly unending variety of styles starting from a simple text prompt. 
Unlike Midjourney, Dall•E can edit existing images by adding or removing elements, thanks to a better understanding of the relationship between the image’s content and the input text. 
Another advantage of Dall•E is the ease of use for people with limited experience. While Midjourney uses the online platform Discord as the main interface for the model, OpenAi’s software can be accessed through a more straightforward online platform.

American startup Stability AI started developing its Stable Diffusion generation model before Midjourney. Stable Diffusion can be accessed through cloud-based services such as Dream Studio, but it’s first and foremost an AI model that can be downloaded, trained, and run on local hardware that’s powerful enough to support it. 
While Midjourney’s results can be steered mostly through fine-tuning the generating text prompt, Stable Diffusion adds many more options to guide the model to the desired result.

Adobe has been working at including AI-powered features in Photoshop, Lightroom, and other software of its Creative Suite for years, thanks to a framework called Adobe Sensei. Recently, the company doubled down on AI image creation by introducing its generative AI model, Firefly. 
Adobe Firefly is still in beta at the time of writing, and users need to sign up for a waiting list to use it. We’ve tried it, and it’s surprisingly good, considering it is Adobe’s first attempt at a standalone generative AI model. What’s most exciting about the development of Firefly is its future integration into Adobe’s software products as an additional tool for graphic designers and illustrators to enrich their works through AI. In our gallery above we’ve asked the three models to generate the same fake picture of the Pope that recently made the rounds. Disclaimer: all images included in our gallery are generated through AI tools.