[Tutorial] Making 2 Specific Characters in the same scene (Patreon)

Hi. In today's tutorial I still didn't have the time to make a video so this one will be images only.

But I've uploaded the images to the image preview of the post.

Let's start with some explanation/disclaimer.

• I'm not going to make a complex sex scene because it would indeed take too much time and it's the same principle that you will learn with 2 girls like this.

• You can use regional prompter / latent couple if you want, I usually prefer to use a position lora/lycoris or controlnet.

• I know it's not perfect and they are not exactly in the anime style. If you want them to be in a specific style just use a style lora since the beginning or use a LoRA that has specific style trained with them.

• This is not a tutorial about how to make the character look the best as it possibly can.

• I've blurred some images so I don't have any trouble with any unknown rule that I'm not aware in the future.

• Be polite and ask your questions without disrespect and I'll gladly answer them.

• I'll not share prompts and settings from my public works.

Okay, the final result is that image which you can download and use the metadata as you want, but it will be useless in terms of prompts.

You can have a better look at the images by looking at the image post instead of the ones I'll post here on text. I'll only post them here so you don't get lost.

—————————————————————

Section 01

—————————————————————

Okay so I've made a simple scene. You can check the prompts. The use of { } are purely for organization. The use of lora with strength 0 is also so I don't get lost because I have hundreds of lora and I don't want to look them up again.

After I've generated 4 images, I've picked the one that I judge that could be used. Notice that my prompts are intentionally close to what the characters look, this is to make the next step easier, but you can have in different colors if you want.

NOTE: You can use one of the characters as base lora if you want to make sure one of those styles are correct so you only need to correct one character.

—————————————————————

Section 02

—————————————————————

Now I've sent the image to the inpainting tab, changed the settings and prompts. My focus now is to make the first girl look like Inda Karane. So I used the prompts fro mthe lora and increased it's strength.

Notice that the lora is more flexible and it's not a very stylized lora, so the girl will not look exactly like in the anime. Also I'm using a base model that is not completely 2D. (hassaku)

You should also notice that I didn't inpaint the hands. The reason for that is that when using a lora, there is a tendency to make worse hands. You can ignore this if you want to increase the likeness even more, but you likely have to fix the hands later.

 These are the settings used for the inpainting. This isn't a rule. The only important parts are the Scale . Inpaint area (Whole picture) and Denoising Strength. The less the character looks like what you want, the higher should be the denoising strength.

—————————————————————

Section 03

—————————————————————

Now let's change the settings and prompts so we can have a better face for her. I've changed the mask blur, the size and the inpaint area to Only masked.

The reason for this is that models in 1.5 are usually trained with 512x512, if you work with big sizes there will be more distortions and my intention here is just to have good eyes.

Notice that the prompts are focused on the character face, so I changed them for that.

So here we go and our Inda is ready. Now we should focus on Hasegawa Chisato.

—————————————————————

Section 04

—————————————————————

Now we have different prompts and settings. I did it like this and noticed a problem. The lora model I'm using is just TOO flexible. So it's not changing the style enough in my opinion. Even using a very strong denoising strength. I've used the same settings as for our first inpainting, the difference is the denoising strength being 0.75 instead of 0.55.

So then I decided to keep that new image and inpaint it using another model. For this model I had to change the trigger word. So pay attention for that. One model was "hasegawa chisato", the second model is "chisato hasegawa". (Yes this makes difference)

Now we have someone that looks a bit more like what we want.

For the purposes of this tutorial I'll end this body inpainting now, but you can try to improve her look as you want, you probably know that at this point it's a matter of prompts and inpainting.

—————————————————————

Section 05

—————————————————————

Remember when I inpainted Inda Karane face? Now we do the same for Chisato. The same settings (512x512, only masked area, denoise 0.35), remember to not screw with these settings or your inpaint will go wrong.

Then we are back to normal inpainting, notice the simple prompt. I'm using the same setting as the first inpainting, but with a higher denoise strength, then we fix the arm and here we go.

I hope this helps you doing 2 characters together. Remember that it will be increasingly more difficult the more complex is the scene.

Note: Sometimes is more worth to make the characters separately and join them in photoshop, then make some inpaintings with low denoise strength to blend them in the scene.