I love emojis!
This might be a cliché, but I consider emojis to be the hieroglyphs of the Internet. In a text-based medium, when measuring character counts, emojis are an efficient way to communicate emotion, ideas and intent. And, they are fun! I love them 😍.
So, when @hafeezhaqq on Threads suggested that “prompting just using emoji works too” for GenAI image generators like Midjourney, I was only too curious to try it out. That was last Sunday.
Midjourney Emoji Week
For the last 7 days, as part of my daily Midjourney Journal, I have been creating images using Midjourney with prompts that are as long as a single emoji. I call it, “Midjourney Emoji Week.”
And the results have been fascinating! It was very interesting to see how Midjourney interpreted these emoji-prompts. Some were out of this world, some were head scratchers, while some were just plain wrong. Regardless, all those generations were aesthetically stunning — something that MJ is known for.
How would Bing do?
But all of this got me curious: how would something like Bing (powered by the new DALL-E 3 model) do with these emoji prompts? Would Bing interpret the emoji-prompts just as well/better than MJ? Would the generated images be just as aesthetically pleasing?
So, on this last day of “Midjourney Emoji Week,” I decided to use Bing to generate images for all 7 emoji-prompts that I used with Midjourney – and then do a compare and contrast.
🐶 🍔 🛺 🚀 🎮 🪐 🌏
On the left/leading column, I present the results by Midjourney; and to the right/trailing column, I give you results by Bing. Before each set of results, I will offer the emoji that I gave as a prompt to both Midjourney and Bing. To either MJ/Bing, I gave no additional qualifiers (e.g., artistic, cartoonish, photo-realistic). I just took the results that GenAI tools gave me. These results, and any artistic flair therein, are the tools’ interpretations of what the emojis mean (in a way).
🐶
Midjourney
MJ understood the emoji. And the images of these dogs are very aesthetically pleasing. None of them are photo-realistic, but that was also not specified in the prompt.
Bing
Bing too understood the emoji. The images are it produces are interestingly photo-realistic (something that DALL-E 3 seems to excel at). And while the dogs in the images are really cute(!!), there is limited variation in the images themselves.
🍔
Midjourney
Again, MJ understood the emoji. These are some very artistic renditions of a hamburger. I personally like them all. I particularly love that three out of 4 of these burgers have teeth — there is a subtle metaphor going on there. Fun! But again, none of them are photo-realistic – it is as if MJ wants to apply artistic styles when not otherwise mentioned (will have to explore this more).
Bing
Bing too understood the emoji. And once again (and unlike MJ), Bing’s generations are photo-realistic. The burgers look yummy! But again, I have to say, they are repetitive in style, with limited variation between them.
🛺
Midjourney
MJ did not understood the “auto rickshaw” emoji. These are some fantastic image generations, but they do not match/fit the prompt at all.
Bing
Bing understood the emoji a little better. But for some reason forgot the “auto” in the auto-rickshaw. But I have to admit, these generated images of rickshaws reminded me of my childhood days in India. You see, I would take these to school 🙂
🚀
Midjourney
I think MJ got the prompt, but it veered more towards astronauts than rockets. All images continue to be jaw-droopingly stunning! And they continue to be artistic renditions.
Bing
Bing certainly understood the emoji, and produced results that stuck to the brief better. All images show a rocket on its way to the heavens. But in true form, not a lot of variation in the images – some, not a lot. I am beginning to think that such results might be useful for use-cases that need mostly the same image, but with subtle changes.
🎮
Midjourney
In my book MJ got the prompt. However, that 2nd image is a bit of a head scratcher (why is it related to gaming?). But MJ more than makes up for it with the other three images — they all show gamers in some state of playing video games. All images continue to be visually stunning, and have an artistic flair!
Special Callout: I am in love (just stunned!! actually) with the first image: If you notice, what is projected on the big screen is a mirror of what is shown in the TV/console in bottom right of the image. I have never seen that level of detail (or semantics) before with generated imagery.
Bing
Bing understood the emoji again. And again, it produced photo realistic images of the object depicted in the emoji: a video game controller. And perhaps this is the thing about Bing/DALL-E3: It seems to stick to the brief really well with these emoji prompts, to produce accurate photo-realistic images, with limited variation between them. That seems to be the default. There is not a lot to complain here to be honest.
🪐
Midjourney
MJ did not understand the emoji. Overall, it seems to be a hit or a miss with its interpretation. But the results are aesthetically pleasing regardless. I do find it funny that MJ interprets the “planet with rings” emoji as “an alien world of balloons, balloons shaped as mushrooms, or just mushrooms.” 🤣🤣🤣
Regardless, MJ at least produces the images of an alien world I guess, which is more than what I can say of Bing…
Bing
Not sure what happened here with Bing. I tried a couple of times, but it kept giving me images for “a mouse face”. This was curious, because before generating the images it spit out a text response that interpreted the 🪐-emoji correctly: “🪐 is the emoji for a ringed planet, which is a type of planet that has a thin, flat disk of dust and debris around it.” Only to create mouse-face images after that. Seems like a bug in Bing’s (chat) system (not necessarily the GenAI itself.)
🌏
Midjourney
I am not sure that MJ understood the emoji for the “Earth Asia”. Overall, these seem to be renditions of alien worlds that have giant elephants with large moons, and potential domes around said moons. I suppose Asia does have elephants. 🐘🤷♂️
Images are superb. Just not what I would have expected for the prompt.
Bing
Bing understood the prompt really well! Not only did it generate images of globes in response, it did so with Asia/Africa in focus. If I were to nitpick:: Ideally, it should have put Asia/Australia in focus. But it is still impressive.
Overall, these results bring up two questions around defaults for me:
- Does MJ have an artistic bias, with short/minimal prompts?
- Is DALL-E3 setup to produce photo-realistic images by default?
I will be exploring such, and other, comparisons between Midjourney and DALL-E3; and perhaps even add StabilityAI’s open source models into the mix.
For now, I will call it an end to this first iteration of “Midjourney Emoji Week”.
— vijay, enjoying a quiet Sunday evening.
Winter Rant 
Leave a comment