Many people are usingMidjourneyWhen generating characters, we basically encounter a problem, that is, it is difficult to maintain the consistency of the character. Today, I will share with you the method of controlling the consistency of the character, I hope it will be useful to you.
The character consistency tutorial is here again!
Preparation:
1. Midjourney account
a. New registered users have 25 free uses
b. The cheapest membership price is $10/month (200 generation quotas)
2. Unable to communicate in Chinese, you need to learn how to write prompt
3. For those who know less about MJ,I suggest you follow my tutorial first.Then study other prompts yourself
First understand two instructions
It doesn’t matter if you don’t understand. Next, we will carry out the actual operation. I will put the prompt of each step in text form for your convenience to copy.
1. Generate pictures
In the dialogue interfaceInput/imagine,ThenInput prompt
|
Create a highly detailed, Pixar 3D render of a game character on a white background, full body shot. The character is a teenager cute girl, wearing black hoodie, and blue jeans |
2. Copy the image link
Selected pictures, directlyRight buttonSelect [Copy link]
3. Change the character's actions
In the dialogue interfaceInput/imagine,ThenInput prompt
This step is important, let's take a look at the prompt disassembly
- The white words in front are the prompts of the first picture (reference picture).Don't change,Copy it directly
- At the end of the previous prompt,Adding Actions【She is waving (She is waving)】
- Then enter[space] [-cref] [space] [paste link]
- Must haveSpace! Otherwise it will go wrong!
Then we get the different actions of the same character
Isn’t it quite simple?
Next, let’s look at other attempts.
4. Real-life style pictures
passprompt to let MJ generate a real person pictureMaintain consistencyThe effect is pretty good
Prompt
|
Reference Image |
photo of an asian woman with black long hair, brown eyes and tan skin. She is wearing a white shirt |
|
Change background |
photo of an asian woman with black long hair, brown eyes and tan skin. Behind her is a futuristic city a night with neon lights. She is wearing a white shirt -cref image link |
|
Change expression |
photo of an asian woman with black long hair, brown eyes and tan skin. She is shocked and surprised. She is wearing a white shirt -cref image link |
The red text is based on the reference picture promptAdded change prompt
But if it isUpload a photoTrying to maintain consistency, but the results are not satisfactory
Foreign stars can at least look a little bit like them (such as the one in the upper right corner)
Let's take a look at the domestic stars, Lao Guo
Not at all
5. Anime style pictures
The consistency of anime style pictures,I personally think it is very goodLet me show you the Sauron I made.
It should be noted that when using MJ to make anime-style pictures, you needSet to Niji mode first
Enter in the dialogue box/settings, Enter to send
You can refer to my settings
Prompt
|
refer to |
Roronoa Zoro |
|
Figure 1 |
Roronoa Zoro wearing a black suit-cref image link-cw 0 |
|
Figure ② |
Roronoa Zoro wearing a pink and blue suit -cref image link -cw 0 |
|
Figure ③ |
Roronoa Zoro wearing a black suit with open shirt -cref image link -cw 0 |
|
Figure ④ |
Roronoa Zoro is waving-cref image link |
I used the character consistency of Zoro.[-cw] command
A comparison of the Promp and the corresponding image should give you an idea of what the [-cw] command is controlling.
- Figure 4, I didn't use [-cw], Midjourney will default to [-cw 100], which means that the character's face, hair, and clothes stay the same, and only make changes to the movement based on my newly given cue word [is waving (在招手)].
- For Figures 1, 2, and 3, I used [-cw 0], which means that Midjourney kept only the face of the reference image, and then made changes to the image based on my newly given cue words.[-cw 0] Give the AI more room to play.
6. Other consistency
Friends in the exchange group mentioned【Other consistency】I also tried
Product Images-Apple Watch
Animal Illustrations-Pixar Style
Animal Pictures-Non-Existent Animals
After a series of tests, Midjourney can now assist us in quickly producing drawings. It can be very competent in some positions with strong repetitiveness and low importance, greatly solving our efficiency problems. However, it is too early to say that it can completely replace designers or painters.
Compared with the character consistency of ChatGPT, Midjourney is obviously much more complicated to operate.But more professional:The generated images have more options and more editable parts
Midjourney finally took an important step forward in terms of character consistency.
Although it mayNot yet fully consistent with standards, but for the 70%~80% task, I think it brings a significant efficiency improvement.
At the same time, unlike Stable Diffusion, Midjourney does not require difficult and costly LoRA self-training, but only requires providing/producing a picture.
Everything will fall into place.