Midjourney powers its AI image editor — here’s how to use it

Midjourney recently announced an update to its image editor, with a more robust and attractive web interface than the older editor that’s still available in Midjourney’s native ecosystem, Discord. This seems to suggest that Midjourney is looking to develop its own interface as a go-to for image generation, similar to what Leonardo and Ideogram are doing.

This feature gives Midjourney an edge over other alternative editors, such as OpenAI’s Dall-E editor, which allows users to inpaint using natural language but does not offer outpainting capabilities. Ideogram also does not offer outpainting or zooming capabilities. In fact, currently the most powerful generative AI editor is Leonardo, but it still lacks some of the features offered in this Midjourney upgrade.

Midjourney has only recently started allowing users to customize their output, essentially training them to reproduce embeddings or use LoRA for a specific style. To do this, users simply select the images they like for direct comparison. Midjourney records all the images the user likes, finds similarities and common elements in each composition, and generates a style that is completely altered in later generations to better suit the user’s preferences.

This is a unique feature not found on other proprietary image generation sites, and an approach that would take a lot of time and resources to achieve in a local model.

Edit in Midjourney

As we mentioned, the new image editor is only available in Midjourney’s web interface: to access it, head over to Midjourney.com, sign in using your Discord or Google account, and click on the “Create” tab to start generating and editing photos.

Once you’re in “Create” mode, enter your prompt in the text box at the top. It will ask you, “What do you want to imagine?”

Once you have filled in the prompts, press the “Enter” key to generate the image.

Once you’ve generated your images within Midjourney’s web UI, you can click on any image from the Create tab to access a menu of editing tools, which will open up a range of options to refine and manipulate the image you’ve created.

Change

The “Vary” feature is similar to the “image-to-image” features offered by other AI art models. Midjourney uses the original image as a base and generates a new set of variations that reference the initial artwork. Selecting the “Subtle” option reduces the strength of the noise reduction, making the new generation match the original image more closely. Conversely, selecting “Strong” increases the noise reduction, allowing the model more freedom to generate images with less influence from the reference.

Luxury

The “Upscale” feature allows you to increase the resolution of your image using generative AI, resulting in a more detailed image. The “Subtle” setting keeps the upscaled image closer to the original, while “Strong” gives your model more room to introduce more elements and changes while still maintaining a connection to the reference.

Remix

The “Remix” option allows you to add variation to your image by changing the prompt, resulting in a more creative result than a simple variation.

Rerun and Edit

Click the Advanced tab and you can select the Redo option, which regenerates the image without affecting previous iterations.

In addition to this, clicking the “Editor” button unlocks Midjourney’s full suite of modern editing tools, as explained in the previous section. This is where all the magic of Midjourney’s new features happens.

Use Midjourney’s powerful image editor

Clicking on the “Editor” button will open a new interface with your image in the center of the screen, from here you can play around with several options and sliders that affect your image.

Here we explain what each option does and how you can use them to improve your image production.

Erase and Restore

One great feature of the Midjourney editor is the “Erase” tool, which allows you to remove or change specific elements in the generated image. Click the “Erase” button to select the area you want to change, then you can enter new prompts and Midjourney will generate updated content for that specific area.

For example, I wanted to generate a party hat, so I selected the “Erase” tool and painted on the robot’s head.

After selecting the parts I wanted to change, I filled in the new prompts specifying the elements I wanted, such as “party hat,” and clicked “Generate” to allow Midjourney to update the image accordingly.

The prompt doesn’t need to be super detailed, at least in terms of style. Midjourney understands the overall aesthetic of the image and generates elements that fit the composition. Here’s what the final image looked like:

Keep in mind that Midjourney recommends selecting an area that’s a bit larger than the specific element you want to modify, so that the AI ​​system has enough context to understand your configuration and generate consistent results. If you select an area that’s too small, Midjourney may struggle to produce a satisfactory output, or may not generate new content at all.

Complementing the Erase tool, the Restore feature allows users to undo an erase and return the image to its original state. This iterative workflow allows users to easily experiment and refine their work, unlocking new levels of creative expression.

Undo, redo, reset

Next to the Erase and Restore tools, the Midjourney editor also provides a set of intuitive icons to manage the iterative editing process. This is pretty self-explanatory, but here’s what each button does:

The Undo button (the back arrow) allows the user to undo the most recent change and return the image to the step before the last edit.

The “Redo” (forward arrow) function does the opposite, undoing the last “Undo” action and reapplying the most recent edit.

The “Reset” button (the two arrows moving in a circular motion) will return the image to its original, unaltered state – essentially reverting the image to the way it was before the user began the editing process.

These simple yet effective controls encourage users to experiment freely, knowing that they can easily undo or reapply changes if necessary.

Scale and aspect ratio

The Midjourney editor also features a “scale” slider that acts as a personalized zoom. By adjusting the scale, users can expand the image canvas and allow Midjourney to generate new content to fill the available space. This is what distinguishes Midjourney from its own competitors that don’t have similar functionality.

For example, a close-up image of a couple watching a soccer game can be zoomed in to include the entire crowded stadium while leaving the couple as the central focus. This zoom differs from traditional “out-painting” techniques in that it doesn’t change the size or resolution of the image itself; instead, by simply magnifying the view, Midjourney allows you to add new elements to the enlarged canvas.

Let’s see this tool in action: I started with a close-up of three robots in a factory and scaled the image down so that it only took up 20% of the canvas space.

After tweaking the prompts and clicking “Generate,” Midjourney imagined a complete factory scene with the same aesthetic, filling in the remaining space.

Please note that when using the “Scale” feature, the artist must provide detailed prompts up front, so that Midjourney has the information it needs to generate high quality, consistent content to fill the enlarged canvas. The end result looked like this:

Complementing the Scale slider, the Midjourney editor also provides a button for adjusting the aspect ratio of your image. This allows you to stretch your image to fit a specific aspect ratio, such as portrait, rather than simply cropping it, and Midjourney will generate the necessary extras for you.

Similar to scale, it’s a good idea to modify the prompt so that the new generation doesn’t miss out on quality. In this example, we converted a widescreen (16:9) image to portrait (3:4). To make it seamless, we modified the prompt and asked Midjourney to generate a table by giving the robot our whole body, and the AI ​​returned the following result:

As with the “Scale” feature, when using the aspect ratio tool, we encourage you to adjust the prompts to ensure the desired quality and aesthetics are maintained in new generations. The combination of scaling and aspect ratio controls allows users to seamlessly adapt Midjourney-generated imagery to a variety of social media formats and layouts.

Brush size and image placement

Finally, the MidJourney editor can move images around within a composition: by hovering over the center of an image, users can click and drag the image to any position on the canvas.

This feature is particularly useful because it lets you preserve elements you want to keep in your image while generating new content to fill in the white space around it. For example, if you have a central close-up of a subject you want to keep, repositioning that central area to one side will allow Midjourney to generate entirely new background elements that complement your main subject.

Also, next to the “Scale” slider is the “Brush Size” slider, which allows you to choose the size of the area selected by the “Erase” or “Restore” tools. So basically, the larger the scale, the larger the area you can paint with one click. Use larger tool sizes to select areas and smaller tool sizes to select elements. This means that if you want to change the nose, for example, a small brush will allow you to select that area without affecting the eyes, while a larger brush will allow you to select the whole face easily.

This UI lets you be as creative as you want, without having to go through each step like in the Discord interface: for example, you can scale the aspect ratio, move the image to a different part of the canvas, erase parts of the original composition, change the prompt to whatever you want, and do all this in a single generation.

So fear not, the tool is very intuitive and no harder than dealing with, say, the ComfyUI workflow. Midjourney is also working on new updates with an eye on Midjourney v7 being released within at least 4 months, so why not have some fun with some image editing while you wait.

Editor: Ryan Ozawa.