Create Reference Sheets with Midjourney Video

Overview (TL;DR)

Thanks to Midjourney Video, we now have a new way of creating consistent turnarounds of our characters! Here’s an overview of the steps:

Generate a simple front-view of your character on a plain background.
Select the Animate Manually option and use Low Motion; describe the movement in your prompt like: girl 3D model turnaround. Optionally, select Loop to reinforce a complete 360 turn.
Extract the still frames you want: Open the video in the full view, right-click image, and select Open Video In New Tab. The video appears in an external player. You can scrub to any frame you want, right-click, and select Save Video Frame As.
Combine extracted frames into a turnaround sheet. </aside>

Wait, you want me to use Midjourney Video to make images?

</aside>

Yes! Even though video is used for making images move, we can use that movement to our advantage and turn them back into still images with a level of consistency not possible before!

Thanks to Midjourney Video, we now have a new way of creating very consistent turnarounds of our characters.

1. Generate the character.

Generate a simple front-view of your character, on a plain simple background.

As an example, here is an example character and prompt!

Straight-on character design of an anime Catgirl wearing a simple white Mayan patterned tshirt, Mayan patterned white shorts, isolated on a grey background --ar 58:77 --niji 6

It is recommended to have your background a simple solid color, with no extra details.

I chose a grey background because I wanted the background to be distinct from the white clothing.

If your character has stray elements, such as props or background elements, it would be a good idea to remove them with the Editor

Once you have the character design you like, you can now prompt the turnaround.

This is where you can find the Remix button after typing /settings in Discord

2. Generate the video turnaround.

When we give the character to Midjourney Video to generate the turnaround, we must use Animate Manually to adjust the prompt.

On Web: Select the Animate Manually option in the animation panel, then select Low in the Motion settings in the prompt bar
On Discord, make sure Remix mode is enabled in your /settings command Then you can click Animate (Low motion). With Remix mode on, this will let you adjust the prompt when animating.

What we will be doing is prompting for a 3D Model, even if your character is a 2D or Realistic style.

You do not need to over describe the character in the video prompt, as longer prompts can get harder to control.

Mayan girl 3D model turnaround --ar 58:77 --motion low --video 1

I chose to prompt “Mayan Girl” (Instead of catgirl) because I did not want the character to have a tail, and “Catgirl” was giving her one.

We chose to use 3D Model in the prompt as modellers often give turnarounds of their models to show them off, and replicating that look helps for a nice smooth turn.

This result was nice and clean, and gave me exactly what I was looking for. Now we can select our still frames to save.

You don't necessarily need the character model to turn around fully if they are symmetrical, a half turnaround (front to back) may result in cleaner details.

Optionally, you can toggle on Loop by clicking the below checkbox on 🌐 Web, or by adding the parameter --end loop on Discord, if you require a full 360 turnaround, such as for asymmetrical characters. Sometimes the results can come out a little more "Jittery" so it might take some patience

3. Extract still frames.

Now we have a video of our turnaround, we are able to pull some frames out to use in our turnaround sheet.

It is recommended to do the following process on the Website.
If you generated on Discord, there will be a link to (Open on the website for full quality) next to the prompt on your video grid.
This technique works for most Chromium based browsers, it may not function the same on Safari, or other non-chromium tools.

1️⃣ Step #1. Open up the video you like in the full view by clicking on it, then right click image and select Open Video In New Tab

2️⃣ Step #2. You are now able to view the video in an external player. Here we can scroll any frame we want, then Right Click it and select **Save Video Frame As**.

4. Build the character reference sheet.

We already have the front view from the original image, so I like to pick a side view, back view and a 3 quarter angle view.

You may want side views of both sides if the characters design is asymmetrical, but with a good turnaround you can pick any angles you may ever want.
You can combine these images together using any image manipulator tool you like, or you can use the Midjourney Editors’ layers function.

😲 A small mistake!

When including this in the sheet, I noticed the background had a slight gradient and was not a solid color. This may not matter to you, it didn’t for me, but if it does just check if a gradient is there before running video generations.

Mayan girl 3D model expression preview. She laughs, then gets angry --ar 58:77 --motion low --video 1

Optional: Add expressions/poses.

You can also use Video to create some expressions to include in your character sheet, or for use in further prompting. I try to get a couple expressions in a single generation to make use of the available time.

You can also prompt for more complex poses that crefs/orefs may struggle to do on its own, and utilise that by both image prompting and a cref/oref

This also allows lower --cw and --ow weights while maintaining character accuracy, since your own character is driving the pose. This improves pose control while not losing the character!

gompasta_Mayan_girl_3D_model._She_pulls_a_martial_arts_pose_a_f22d5d1a-2b3f-45c4-8fcb-34478ea277c4_1.gif

gompasta_Mayan_girl_3D_model._She_pulls_a_martial_arts_pose_l_f1992c3a-2ced-45d3-b2b3-d372f26c3e83_0.gif

</aside>

Nope! these are not limited to anime characters, though that was my example. This can be done in any style, be it photographic, 3D or for products, though depending on the style of image used, these may require a high motion mode (which may mean more unwanted character movement).