Midjourney is an AI tool that leads to the door of limitless creativity, transforming human imagination into visual art. Only this time, we’ll unlock it not with words but with IMAGES. Yes, you’ve read that right!
An image can be used as a prompt on Midjourney, serving as a reference to the art it will generate. All you have to do is combine them with different photos, texts, and other elements; you name it. You can even get more creative outputs with a bit of thinking outside the box.
Don’t worry; it’ll be an easy task.
The 3D photorealistic version of Luffy from One Piece, generated using a similar 2D photo as a reference, is a prime example of the visually stunning results that can be achieved through image prompting. But this is only one of the endless possibilities that can be accomplished with this technique, so let your imagination run wild!
This guide will walk you through all the simple yet advanced Midjourney techniques for creating next-level AI-generated art using images.
How Does Image Prompting Work?
Midjourney allows you to upload one (at least two if no text prompt) or more pictures that’ll serve as a reference for the AI in creating a new output. Note that the result will have similar attributes to the source images, but they won’t look exactly like the subjects in your photographs. That said, consistent features or characters can be accomplished with patience and a lot of trial and error.
Take the example below. Replicating human features is the most challenging in my experience.
This one, in particular, took a few rerolls in order to get right and I had to tweak the prompt a couple of times since the kimono prompt consistently resulted in an Asian-looking woman at first. Granted, it’s not my best work, but I wanted to show you something I did from scratch.
The time it takes to get the desired output can be anywhere between 3-40 minutes depending on the complexity of the design and your level of tolerance for imperfections.
Image + Text Prompts
Stop wasting time and let’s do this. I assume you already know the basics of Midjourney. If not, check out our 10-Minute Guide to Using Midjourney.
Let’s start by letting you generate an image out of a basic image and text prompt one-time, and then we’ll proceed to more high-level techniques.
Log In to Discord
Midjourney currently runs on Discord, meaning you must create an account first. Once you have it, visit the official channel of Midjourney by clicking Join the Beta or Sign In if you don’t have an account.
Wait for a while until the page directs to the channel. You’re free to navigate, but if you want to try generating images immediately, enter any available Newcomer Rooms.
Find a Reference Image
There are two ways to use an image prompt in Midjourney. Either you upload your image on Discord or copy one from the web. Note that you only need to do one of these two steps.
Using Images Via Web
Midjourney supports images via web URLs from Google and other sites; you just need to paste the link in the text field.
Note that the URL needs to end with either .jpg, .png, or .gif. For example:
https://cdn.britannica.com/78/43678-050-F4DC8D93/Starry-Night-canvas-Vincent-van-Gogh-New-1889.jpg. Other formats are not supported.
Uploading Your Own
If you want to upload your own photo, simply drag the file to any of the newcomer channels and click Enter. Just like this:
Once the upload is done. Click on the image to enlarge it. Then right-click and choose Copy Link. This URL can now be used as a prompt.
Generate an Image Using a URL
Type in the text field: /imagine [URL/s of image/s] [text prompt (optional)]
IMPORTANT: If you include a text prompt, then MJ can generate with just one image. But without text, MJ v4 requires at least two images to work with.
In my example, I typed in: /imagine [URL of Starry Night painting] guy smiling
And the result:
As you can see you can get pretty decent results right off the bat. However, if you want your image to look a certain way, you’re gonna need to learn a few advanced techniques made easier with our intuitive step-by-step process.
Next step, try image-only combinations.
Image-Only Combinations
You need to use at least two images as prompts if there is no text, but that’s only the minimum. You can actually combine up to 5 photos.
For this example, I’ll be using three: a cyberpunk wallpaper, a photo of Jennie from Blackpink, and Athena the Greek goddess. These are just random jpegs taken from the web. I didn’t really visualize or plan out stuff. I just want you to see the results if you do this without care.
Below is what the prompt will look like. Just the links without any text.
Alternatively, if you don’t like using URLs and want to work fast, you can use the /blend command, which is faster.
Simply type /blend on the Discord text field and upload as many images as you want. It’ll generate the same result as the URL method.
The final output clearly doesn’t look good, but that’s okay because now you know how important it is to choose the right set of images. In this case, I didn’t even take into account that two of the images have a horizontal orientation while one is a portrait, yet I still use the default square aspect ratio.
Advanced tip: You can set the dimensions of the final image by simply adding “–ar [insert aspect ratio]” at the end of the prompt. Midjourney supports most aspect ratios, such as 1:2, 2:1, 2:3, 3:2, 4:5, 5:4, etc. More info on aspect ratios here.
The resulting pictures can either be predictable or surprising, depending on the complexity of the inputs. It’s also possible for the AI to generate an output that doesn’t match your standards, but there are some ways to improve and get the results you want.
Remixing and Rerolling
If you’re not satisfied with the results, you can continue to reroll the image, which essentially means letting Midjourney run it again and generate other variations. For that, you need to enable Remix mode first.
Activate Remix Mode
Simply type: /prefer remix in any text field on MJ’s Discord to activate.
With Remix Mode, you can effortlessly modify prompts, aspect ratios, parameters, etc. to create new and unique variations of your starting image. This mode utilizes the essential elements of your initial image to produce a fresh and altered result.
Let’s Reroll
In this instance, I’m going to reroll the crappy image of Jennie I generated earlier to make it look beautiful.
Obviously, I’m gonna need to change the aspect ratio to match it with the generated image.
I think the top-right output is the most workable for me, so I’ll reroll that one. The key is to reroll the image that’s closest to your desired result. So I’m going to click V2 and a text prompt:
Eleven minutes and eight rerolls later:
Ultimately, you’ll probably get the result you’re hoping for if you try hard enough.
Just remember that while Midjourney can undoubtedly transform text and image prompts to greater heights, it may sometimes fall short of expectations, as it’s still changing and improving. Sometimes I have to do 30 rerolls, edit on Photoshop, then upload again to reroll.
Don’t let anyone tell you this is not art. It isn’t as simple as the haters make it out to be. Hone your craft as this is the future!
Who knows what incredible advancements it can bring? Join
You might also like:
- 16 Innovative Use Cases of AI-Generated Art in 2023
- AI and Art: Opening Doors to New Realms of Creativity
Join our newsletter as we build a community of AI and web3 pioneers.
The next 3-5 years is when new industry titans will emerge, and we want you to be one of them.
Benefits include:
- Receive updates on the most significant trends
- Receive crucial insights that will help you stay ahead in the tech world
- The chance to be part of our OG community, which will have exclusive membership perks
