Midjourey: Multiple Character Workflow + Training Runway Models.

Fam,

I have to say something…

This community we’re building is so damn cool.

I would have never expected robot art would grow into something we could consistently converse about week after week.

You all make this fun…and I can’t thank you enough.

(Sappy intro…over)

That being said…here’s what rapping about this week.

1. Midjourney Mastery Course: Tier (1) Sold Out.

2. Multiple Consistent Characters with CREF.

3. Using Your Own Image for Character Sheets.

4. Tips for Training Models in Runway.ml 

5. Midjourney Office Hours.

Let’s boogie.

Midjourney Mastery Course: Tier (1) Pricing SOLD OUT.

Tier (1) pricing for the course is now SOLD OUT.

There are (30) slots left for Tier (2) pricing ($167).

(Before the price increases)

Kinks are being ironed out…

And the first additions will be made to the course in the next 10 days. More videos…more documents…it never ends lol.

Here’s more details:

• What Does it Include: 

  • (67) Video Tutorials.

  • (34) PDF Guides.

  • (14) Modules - Everything from Signup to Storyboards.

  • Monthly Content Updates.

  • (4) Cheatsheets.

  • 7-Hours worth of video content.

Your BETA access price is $167 USD is available for the next (30) people.

(This will eventually be $227)

Now that that’s out of the way…let’s get into it.

Multiple Consistent Characters.

This has been an issue for a while…and Midjourney hasn’t provided a solution.

So we’ll hack around it. (For now)

If you missed it…

Here’s the Process:

1. Build Character Sheets.

2. Use Multiple Character Sheets + CREF.

3. Use Structured Text Prompts.

4. Use Vary Region For Edits.

Character Sheets:

We’ve talked about this here, but they’re important…they give Midjourney more face and body data.

More data = better results.

Keep that in mind…because this will work for other “objects”…not just people.

Quick Character Sheet Prompt Formula:

character sheet, split into 4 photos, shot from multiple angles, Subject: [character description], Wearing: [clothes], Setting: studio, Image type: 70mm portrait, high precision --ar 1:1

Now…once you’re able to create character sheets…this becomes a lot easier.

You want to have multiple facial angles and poses…think about it like you’re creating a 3D model of this character…so they are easily placed in a scene.

Once you have enough data…you can start to build them into scenes…BUT…

Prompt structure becomes important…

Here’s one I’ve been playing with:

Photo type: [shot type] , 2 characters [action], Character 1: [details/wearing], Character 2: [details/wearing], Subject Focus: mid ground, sharp focus, Setting: [environment], Lighting: [light description], Image Type: [tech details]--ar 3:2 --cw 75 --cref [Man - URL 1][Man - url 2] [Woman - URL 1] [Woman - url 2]

It’s not always perfect.

Midjourney tends to “mashup” faces when you give it multiple characters.

It can look like nightmare fuel…or be somewhat funny…either way…that’s not going to work for us.

See example below:

Now…if this is happening to you…try utilizing Vary Region to fix the faces.

Here’s the steps to take:

Step 1: Highlight the face you want to change.

Step 2: Remove CREFs with unwanted face.

Step 3: Remove other character details from text prompt.

(Full breakdown in the Linkedin Post)

It’s not perfect…but it works.

Test it out…and we’ll keep trying to hack the system.

I’ll be adding videos about this to Midjourney Mastery Course.

Making Your Own Avatars with Character Sheets.

I’ve been getting this question a lot.

“Can you use pictures of yourself with CREF”

Short Answer: Yes.

More Detailed Answer: It’s not as straightforward as it seems.

However, the same process can apply.

By utilizing the same character sheet prompt from above.

The catch?…You have to build your own character sheet first.

If you want to inject yourself into Midjourney…there’s a few things you need to know…

If you’re uploading pics of yourself…they need to be high-res.

If they’re not high-res…utilize something like Magnific to upscale them…the lower the resolution the worse the translation will be.

Now, I’m still having the best success with CREF using Character Sheets.

Therefore…make a character sheet of yourself…using different angles and poses.

You will have to make this on platform like Figma, Canva, etc. to create them.

Here’s the one I made of myself:

Here’s the Process:

1. Create a Character Sheet of Your Images.

2. Use that Character Sheet to make more Character Sheets.

3. Use the best ones to place yourself in scenes.

This process takes a few tries.

(Because you’ll be overly critical of your every feature lol)

Quick Tips:

1. Try generating 3-4 character sheets (that look most like you)

2. Sometimes using a good character sheet as an SREF + CREF can help when trying to create more cohesive CS’s.

3. Test using more detailed prompts…and prompts that are extremely minimal.

4. Use character weight to balance.

Training Models on Characters:

If you’re not thinking about it already…this is where all this “Character consistency” is going…

Training models on your characters...that’s where it’s going.

So you can put your character anywhere.

If you’re new to model training…and want to play around experimentally…Runway.ml allows you to train character models with minimal expertise.

Or if you want to learn how to do it with Stable Diffusion…I came across a Good Article to Get Started.

Now, you’ll need a paid subscription to Runway to try this out.

I spoke with Timmy Highley (Community Manager) at Runway…and he had a few suggestions for training your character model.

1. Use 20+ photos of the same character.

2. Use multiple face angles and poses.

3. Use 1:1 aspect ratio for input images.

4. Don’t use complex backgrounds (studio/blank background works best)

5. Use Higher-Res photos (if you have them)

6. If you’re generating the characters (upscale them to make them higher res)

The Process is pretty easy:

1. Go to runway.ml

2. Select “Ai Training”

3. Select “Train a New Model”

4. Select “Character Generator”

5. Upload your images.

6. Train your model.

Once that’s done…you utilize the text-to-image generator to create your character in multiple settings.

Runways image generator is very user-friendly with several preset image styles. So you can play around with different looks.

(This is not a paid promo btw)

I’m just digging how easy they make it.

Once you have some images you like…you can take them into the GEN-2 (motion generator) and start to make animations.

Personal Opinion: Images generated in Runway tend to animate better than images generated in Midjourney…idk what it is…but that’s just my feelings.

Play around with it…have fun.

Midjourney Office Hours:

Nothing revolutionary…but some interesting updates.

1. Social Creation Tools:

   • Rolling out new social features for top users with early positive feedback.

   • Starting small for better control, then expanding so everyone can create their own social spaces.

2. Image Personalization Update:

   • Shifting focus to user-specific preferences, with a unanimous preference for personalized results.

   • Continuous improvements planned with an opt-in feature for sharing personalizations.

3. Tech Enhancements in Version 6.1:

   • Enhancing representation of hands, bodies, and text, and reducing pixel flaws.

   • Beta version 6.0 paving the way for a robust Version 6.1 with speed boosts.

4. Future Roadmap for Version 7:

   • Aiming for major aesthetic improvements without compromise.

   • Expect better prompts, clarity, style, and resolution in V7.

   • Anime video generation is also in the pipeline, with a focus on easier execution than photographic video.

Have fun.

If you need help…DM me on Linkedin.

OR respond to this email with “HELP”…and we’ll chat.

Until next time…

RF