Are you tired of browsing through countless stock images that fail to capture the exact image you have in mind? Well, fear not, as AI models are here to help!
In this article, I explore the results generated by three different AI image generators – Midjourney, Stable Diffusion, and Microsoft’s Bing Image Generator – when given prompts that range from a simple concept to a detailed description.
From a swimming astronaut in the middle of a desert to a cute monster driving in outer space while eating a cupcake, the results are varied and intriguing. So, join us on this journey to discover which AI model comes out on top!
Midjourney vs Stable Diffusion vs Image Creator
Before we dive into the comparison, let’s briefly look at each of these tools and who are its creators.
Midjourney is a small self-funded research lab with a focus on design, human infrastructure, and AI. The team consists of 11 full-time staff members and a set of advisors. The lab aims to explore new mediums of thought and expand the imaginative powers of the human species. Currently if you want to try their tool, you need to create a Discord account and experiment at one of the #newbie channels to create images.
Stable Diffusion the text-to-image model from StabilityAI. It uses Stable Diffusion model which is a type of deep learning model that belongs to the class of diffusion models. These models are designed to generate new data that is similar to what they have learned during training. In the case of Stable Diffusion, the data that it learns are images.
Image Creator is the text-to-image tools created by Microsoft Bing. In the back end it used OpenAI’s latest DALL-E model. It is currently integrated with Microsoft Edge. And all the images that it creates, have a Bing logo at the bottom left.
Alright, now we know our tools, now let’s dig into the real battle!
Prompt 1 – astronaut swimming in a pool in the middle of a desert
I first tested it with Midjourney on the above prompt. Here is the result it generated.
The images created by Midjourney were quite cool to say the least. I quite liked the first one personally given the cool shows and effects it produced. However, Midjourney took it quite literally in the second image where the astronaut was swimming in the middle of the pool which again was in the middle of the desert! The effect of wind blowing in the third one which covered the swimming pool was quite interesting. However, the last one did not depict an astronaut swimming in a pool. So overall, Midjourney did an interesting job of creating the image I had in mind.
Next I tested it with Stable Diffusion.
Stable Diffusion on the other hand had quite different results. Here, the features were quite literal. Like the desert and the swimming pool are quite similar to what we see in real life. This however does not match to our expectations as the prompt is itself quite unrealistic so and unrealistic background was expected. Also, one cannot associate the astronaut in the first and last images with an actual astronaut. So, I’d say, the results generated were not at par with what Midjourney generated.
Next I tried out Microsoft Bing’s Image Creator and checked out the results it produced with the same prompt.
The results were mesmerizing. I actually liked all the images generated by Image Creator. In fact, these were better than the ones generated by Midjourney in my opinion.
But it is unfair to test these platforms on just a single prompt. So I decided to test it on another prompt. A slightly longer prompt with more information.
Prompt 2 – a cute purple monster driving a red car in outer space while eating a cupcake in right hand in cyberpunk style
Here is the result generated by Midjourney.
Midjourney created a cute monster but most of the cars were not actually red in colour which was a specific requirement in the prompt.
Stable Diffusion on the other hand produced this.
Here, Stable Diffusion separated the descriptions all together. In most of the images, the red car, the monster, and the cupcake were mutually exclusive. None matched the prompt I had provided.
Finally, I tested it on Image Creator.
The results were again quite satisfactory. A monster riding a red car in space eating a cupcake was quite accurately depicted in all of the images
Conclusion
In our opinion, Microsoft’s Bing Image Generator generated quite compelling results matching the prompts that I provided. The only drawback being the small Bing logo being added to the bottom left of the images. However, even Midjourney performed well for certain prompts.
So the verdict is out there. Your opinion can differ from ours and why should it not! We would love to hear them in the comments section below. But these opinions we out first impressions with these tools. Expect more in depth hacks and workarounds as we explore these tools. Stay tuned!