How to Use Midjourney Inpainting, Vary Region Feature - Tutorial Guide

  1. Introduction to Midjourney Inpainting

  2. How to Use Midjourney’s Inpainting Feature, Vary (Region): Step-by-Step

  3. Example Use Cases for Midjourney’s Inpainting Feature

    1. Change character attributes

    2. Correct or change hands, arms, and legs

    3. Remove objects from images

    4. Add objects to images

    5. Limitations for Midjourney's Inpainting Feature

    6. Summary

Introduction to Midjourney Inpainting

The long-awaited Midjourney inpainting feature is here!

Midjourney’s powerful, new inpainting feature, which they call “Vary (Region)” allows you to make changes to specific regions of an upscaled image. For those familiar with Photoshop's Generative Fill feature, the concept behind Midjourney's inpainting, Vary (Region), is similar --select a region, give it prompt text, generate results. In my opinion, Midjourney's results are MUCH better though.

Having an inpainting feature available in Midjourney is extremely beneficial for users that do not have access to programs like Photoshop, Gimp, Affinity, etc for making small changes to images.

Inpainting allows you to make those changes as part of your Midjourney workflow in Discord.

Through a new feature in Discord, you can draw around the region of your image that you want Midjourney to "re-roll" or create variations on. Unselected parts of the image will stay the same. You can even change the prompt text to tell Midjourney what you want to see in your selected region.

For example, I prompted /imagine cute dog --v 5.2 --ar 2:3 and got the starting image shown below. Then, I selected the region on and above the dog's head. I removed "cute dog" from the prompt text and replaced it with "hat", and then submitted the prompt. Midjourney gave me a grid of 4 image results and I upscaled one of them.

Midjourney inpainting is available to use in ALL versions of Midjourney!

I was really excited to see that inpainting is available across ALL Midjourney versions.

One important thing to note, however, is that the algorithm for inpainting is different from the algorithms Midjourney employs to create your starting images.

What do I mean by that?

Let’s say you create an image with Midjourney v3 and select a region to "inpaint" or "vary". When Midjourney runs the inpainting job, it will not use the v3 algorithm on the region you selected. The inpainting algorithm is separate from the algorithms that generate the starting images. If you inpaint a region of a v3 image, what shows up in that region will be more similar to what you’d see in v5+.

In the example below, you can see that the style of the castle after inpainting is quite different from the original castle. The prompt text for inpainting was the same as the prompt text used to generate the starting image.

As you can imagine, there is a wide array of scenarios where the new inpainting feature can be utilized.

In this article, I will walk through exactly how to gain access and use the new Vary (Region) / inpainting feature and will share several examples to show just how powerful it is.

Also! If you are interested in some video tutorials, I highly recommend checking out my new YouTube channel where I will be sharing a lot of AI image & video related tutorials.

How to Use Midjourney’s Inpainting Feature, Vary (Region): Step-by-Step

1. Check that you have the new Vary (Region) button

First, you will need to check that you have the new Vary (Region) button. To do this, look below any upscaled image in Discord. If you do not see a Vary (Region) button, you will need to update Discord.

To update Discord on the Desktop app, simply open Discord and press Ctrl+r. The Discord window will blank out momentarily and when it comes back, you should have the button below your upscaled image.

2. Enable Remix Mode

Next, you will want to enable Remix Mode. This will allow you to change the prompt text for the region that you select. You can do this by typing /settings and selecting Remix Mode.

3. Create and upscale a Midjourney image

Now that you are all set up, submit an imagine prompt to Midjourney. For this example, I submitted the prompt /imagine man walking down a city street, close-up shot, daytime, photorealistic --v 5.2 --stylize 250.

Midjourney returned a grid of 4 images and I clicked a "U button" below the grid to upscale one of the images.

4. Click Vary (Region)

Click the Vary (Region) button beneath your upscaled image to start the inpainting process.

5. Select the region you want to inpaint and submit

After clicking the Vary (Region) button, you will get a pop-up window like the one below. There are two selection tools, a box and a lasso. You can select the one that works best for you.

Select the region of the image that you want to change. It is best to select 20-50% of the image so that Midjourney has a large enough subject and context to work with.

After selecting your region, you can change the prompt text at the bottom to tell Midjourney what you want in this region. Then click the arrow button to submit it!

6. Review your results

For each inpainting job you submit, Midjourney will give you the usual grid of 4 and you can choose to upscale one, make variations on one, or re-roll the whole job. You can also go back to your starting upscaled image, click Vary (Region) again and modify your selection. Discord caches the selection region.

Below are a few of the results that I got when changing this man's coat. The text used for the inpainting prompt submission is shown below each image.

Midjourney inpainting results

blue coat, cubism --stylize 250

furry white coat --stylize 250

Midjourney inpainting results

tuxedo --stylize 250

Midjourney inpainting results

blue coat --stylize 250

Example Use Cases for Midjourney’s Inpainting Feature

Change character attributes

I think one of the most popular use cases for Midjourney's inpainting feature will be to change character attributes. These can be things such as hair color, facial expressions, and clothing. Some of this, I demonstrated above, but here are a couple more examples.

Example 1:

I started with the image on the left of a woman doctor. The prompt for this image was /imagine doctor, she, long hair, illustration --v 5.2 --stylize 40 --ar 3:2. Then, I ran 4 successive inpainting jobs. First, I selected here hair and prompted blond. Then I upscaled one of the images of the blond doctor. From here, I selected an area in front of her and prompted clipboard. Again, I upscaled one of the results. Then I selected around her eyes and prompted sunglasses. Lastly, I selected her lab coat and prompted blue lab coat. The final image is shown on the right.

Midjourney image of a woman doctor

Starting image

Result after 4 rounds of inpainting

Example 2:

If you love creating turnaround images of characters, you'll be happy to know that Midjourney's inpainting feature also works on these character turnaround images!

I started with the image on the left by prompting /imagine winter, she, full body turnaround --niji. On my upscaled image, I selected the area around her coat, across all 3 poses and prompted for blue coat, she, full body turnaround --niji. Then, I upscaled one of those results and selected her mid-section and prompted holding a coffee cup, she, full body turnaround --niji. The final result is the image on the right.

Midjourney full body turnaround image before inpainting

Starting image

Result after 2 rounds of inpainting

Fix or change hands, arms, and legs

Of course, if you are prompting a lot of images that have hands, arms, or legs in them, inpainting is a great addition to the Midjourney toolbox. If you get a mangled hand, you can just select the region around the hand and have Midjourney try again.

Below, I added a sword (sort of) to a woman's hand.

First, I prompted /imagine full body pose, she, vintage film, light leaks --v 5.2 --stylize 250 --ar 4:5 --chaos 15. Then I did a 1.5X zoom out (read about zoom out here). On the upscaled, 1.5X zoomed out image, I clicked Vary (Region). I selected an area around her lower arm, all the way down to the bottom edge of the image. Then I prompted holding a sword, full body pose, she, vintage film, light leaks --v 5.2 --stylize 250 --ar 4:5 --chaos 15. Note: it can be helpful to keep image style words in your inpainting prompt (e.g. vintage film, light leaks).

One of the results was a sword, but I chose to upscale the light saber sword 🙂.

Midjourney vintage film image before inpainting

Starting image

Result after adding a "sword"

Remove objects from images

You can remove objects from Midjourney images using inpainting. You can do this with small objects or large objects. If you are removing a small object, it can be helpful to select a larger region around the object.

To demonstrate, I prompted /imagine sad boy, illustration --v 5.2 --stylize 35 --chaos 15 and upscaled one of the results. Then, I selected the region around the boy and changed the prompt text to brown paper notes on a green wall, illustration --v 5.2 --stylize 35 --chaos 15. Below, in the image on the right, you can see that inpainting was able to remove the boy from the image.

Midjourney sad boy image before inpainting

Starting image

After using inpainting to remove the boy

Add objects to images

Earlier, I showed that you can change character attributes such as coat style and hair color. The "frame" for those types of changes already exists in the image. The person is in the image. They already have hair. They already have a coat. It is easy for Midjourney to determine the context.

But you can also add completely new things to your images. For example, I prompted for the mountain landscape shown below with /imagine fantasy mountain range, pencil drawing --ar 3:2 --stylize 250 --chaos 15. Then I selected a large region on the left side of the image and changed the prompt to scottish castle ruins, pencil drawing --stylize 250 --chaos 15 --ar 3:2.

Midjourney did not have a building as context in the original image. However, an old Scottish castle among a beautiful mountain landscape seems plausible, so it was able to create it without issue.

Midjourney mountain landscape image before inpainting

Starting image

Result after adding Scottish castle ruins

Limitations for Midjourney's Inpainting Feature

Works best when larger regions are selected

The inpainting feature works best if you select at least 20-50% of the image. You can change smaller regions, but Midjourney will have an easier time making the intended change if you select a larger region to start with.

Also, you cannot zoom in to get really precise selection. The feature isn't really meant for precise selection. It needs a larger sample space to work with.

Can be difficult to add features that do not match the scene or context

Above, I showed you how I was able to add a Scottish castle to a mountain landscape. This was easy for Midjourney to do because it seems like a plausible image.

Inpainting doesn't work as well if you are trying to add objects that seem out of place for the scene.

For example, I prompted /imagine desert landscape --stylize 250 --ar 3:2 and upscaled the first image below. Then, I used Vary (Region) to select a large region in the middle of the image. I changed the prompt text to whale --stylize 250 --ar 3:2.....and got the result shown in the 2nd image.

Midjourney image of a desert landscape

Starting desert landscape image

Result when I attempted to add a whale. Reminds me of the movie Tremors.

Inpainting / Vary (Region) can only be used on Midjourney images

Vary (Region) can only be used on images created with Midjourney.

I can imagine that this might disappoint some people. You cannot bring in an external image and use the inpainting feature on it. However, you CAN use an external image in a Midjourney prompt. You could create a new Midjourney image using a reference image and inpaint that result, but I know it's not going to be identical to your external image.

If you want to make inpainting-style changes to an external image, you'll need to use something like Photoshop's Generative Fill.

Summary

In summary, the new Midjourney inpainting / Vary (Region) feature is a great addition to the Midjourney toolkit. It allows you to directly make changes and corrections to your image using a Midjourney algorithm. You can change character attributes, add or remove objects from scenes, and build entirely new scenes. The possibilities are endless!

If you'd prefer to watch moving imagery instead of reading, I highly recommend subscribing to my new YouTube channel!

Thanks for reading!

Previous
Previous

Midjourney v6 Alpha is Here!

Next
Next

Batch Upscale Images with Topaz Gigapixel AI: Step-by-Step Guide