I think this is a Lora, so you can add it in stable-diffusion-webui\models\Lora folder. Then you select it by pressing the "Show/hide extra networks" button (looks like a painting with a red sky, white sun, and black ground), underneath the big "Generate" button on the Stable Diffusion page (where you generate all the pictures). From there when you select it, it will appear as text in your prompts. Don't forget to use specific prompts to help it be more specific

Also, OP posted this, so you should probably listen to them:

LyCORIS is a derivative class of LoRA and requires the installation of the LoCon Extension to use. Once the LoCon Extension is installed, you can use LyCORIS just like LoRA, and they will be displayed in the prompt as <lora:MultipleGirlsGroup:1>. However, they are still LyCORIS.

If you want to use LyCORIS, you can refer to Civitai's tutorial:
https://github.com/civitai/civitai/wiki/How-to-use-models#locon

Upscaling usually helps as it will provide more area to generate

upscale, set codeformer and gfpgan. the more the face needs to be fixed the higher the level of the two you need

Here is a list of resolutions in different aspect ratios that tend to work in general (not with this though).

512x512 = 1:1

512x256 2:1

256x512 1:2

704x384 ~16:9

384x704 ~9:16

640x448 ~4:3

448x640 ~3:4

Playing with these some, I've had success with this lora and 768x640. Not perfect yet. I'll update if I keep playing with this and find a better resolution. Essentially, I believe SD is so focused on 1 and sometimes 2 people, that it knows to put more people when you ask, but it doesn't know how to convert those multiple 512x512 ideas onto the pallet correctly.

Hello. I had the same issue as you, I think, but I discovered that my vae was causing the faces to generate oddly. Changing it/turning it off both resulted in better faces.

Overall_Detail and Face_Detail are adjustable embeddings that add extra detail, and lora more_details is a really good detail adjustment tool. The face detail embedding will insist on trying to make portraits, but you can prompt that out and it keeps the extra detail on the face

i think this
https://civitai.com/models/18427/fantexiv08?modelVersionId=24196

@ninkads thanks a lot!!

It will call the right file even if harem_4 appears. And the file name is multipleGirlsGroup_v10, so you can write <lora:multipleGirlsGroup_v10:1> (or <lyco:multipleGirlsGroup_v10:1> depending on how your things are set)

Just concern where you get that harem lora

I don't think the lycoris was trained on 2girl or 3girl photos.

you could probably reduce the pixel width or increase the height

@inbox4karl308 I tried to reduce the resolution, e.g., to 512x768, the girls came into the front or back with distorted faces. Looks like the number of girls happens randomly.

strange, for me the number i prompt gives me one less in most cases. 3girls=2girls, 4girls=3girls, etc

@inbox4karl308 Indeed, in testing with 3girls, I went to 1024 x1024 and higher and ended up with a synchronized swimming team with a dozen women, HAHAHA.

This lora is a lot of fun, but it can give some pretty wild results, most of which are fails. With 50 generations, after clones, crowds, mutations, I'm probably left with 5-10 worth anything. Which is fine, I wish it was higher, but it's not, and I got 5-10 images!

3:2 or 4:3 landscape ratios seem to work best. Even non-standard resolutions can work. Use adetailer to fix faces.

我也想知道。特别是不同的样貌，我目前只能先文生图，然后生成的图把其中一张头涂了重绘，用lora或者roop

@samcassel 用tiled diffusion分区域绘画，每个区域的人物用各自的Lora，就可以实现多人多样貌。

@fan052381 多谢

@fan052381 Are you operating in Tu Sheng Tu? Or what plugin is it?

If you can't use inpainting with ControlNet, I don't see an easy way unless you use some extension like Latent Couple but for selected regions.

Regional Prompter is good for this, look for it as available in the Extensions tab :)