Still pretty new here, like almost three weeks ago. I've been trying to make human male/female anthro pics using most of the Pony XL models. However, many of them end up with just human on human action, sometimes either male or female having only partial anthro features, and definitely either wrong positions when I wanted them to do something else, like generated images show cowgirl despite wanting missionary, but it's the failure of not being able to have human male on anthro female sex without it being the male's POV. Tips when promtting?

masterbeaster50 said:
Still pretty new here, like almost three weeks ago. I've been trying to make human male/female anthro pics using most of the Pony XL models. However, many of them end up with just human on human action, sometimes either male or female having only partial anthro features, and definitely either wrong positions when I wanted them to do something else, like generated images show cowgirl despite wanting missionary, but it's the failure of not being able to have human male on anthro female sex without it being the male's POV. Tips when promtting?

The prompt isn't really what is relevant here. Some say you should separate the characters with the "BREAK" keyword. But that is very ineffective. You have two real options to create duos:
1. Make a sketch (it can be VERY primitive!) and feed it into img2img.
2. Use inpaint (with only masked) to turn one of the humans into an anthro. (and prompt as if you would only want to create an anthro)

I don't like Pony family, so I used usual SDXL model for that quick test. Used "yiffymix v52XL". Sometimes generate wrong combination, like anthro penetrating human or anthro penetrating anthro

How it looks like https://ibb.co/3kdVqYg

For prompt I used e621 tag style

Positive prompt
(human_penetrating_anthro:1.5), duo, penetration, female dragon, male human, doggystyle

Negative prompt
human_penetrated

silvicultor said:
The prompt isn't really what is relevant here. Some say you should separate the characters with the "BREAK" keyword. But that is very ineffective. You have two real options to create duos:
1. Make a sketch (it can be VERY primitive!) and feed it into img2img.
2. Use inpaint (with only masked) to turn one of the humans into an anthro. (and prompt as if you would only want to create an anthro)

Thanks I'll try it out

yetanotheraiuser said:
I don't like Pony family, so I used usual SDXL model for that quick test. Used "yiffymix v52XL". Sometimes generate wrong combination, like anthro penetrating human or anthro penetrating anthro

How it looks like https://ibb.co/3kdVqYg

For prompt I used e621 tag style

Positive prompt
(human_penetrating_anthro:1.5), duo, penetration, female dragon, male human, doggystyle

Negative prompt
human_penetrated

Thanks, but is there a way to also prompt specifics for each character? For example I want the guy to be lean, or a twink, while also wanting the anthro to be a fox (my preference) with specific fur or hair color without those traits messing up one another?

masterbeaster50 said:
Thanks, but is there a way to also prompt specifics for each character? For example I want the guy to be lean, or a twink, while also wanting the anthro to be a fox (my preference) with specific fur or hair color without those traits messing up one another?

In theory, yes. That's the (ineffective) method with the BREAK keyword:

general tags BREAK character #1 tags BREAK character #2 tags

It works to some extent. But really getting what you want would be like winning the lottery. There will always be "bleed over": If you want a human on a deer girl, the human might have antlers, too...
But you can use this method for a start, if you don't want to sketch something first. Than refine the best result with inpaint and change the prompt accordingly. For example, if you have a human with antlers: Only mask the humans head and put "antlers" in the negative prompt. Fix all the errors step by step... this can take some time! Creating nice duos is hard work!

silvicultor said:
In theory, yes. That's the (ineffective) method with the BREAK keyword:

general tags BREAK character #1 tags BREAK character #2 tags

It works to some extent. But really getting what you want would be like winning the lottery. There will always be "bleed over": If you want a human on a deer girl, the human might have antlers, too...
But you can use this method for a start, if you don't want to sketch something first. Than refine the best result with inpaint and change the prompt accordingly. For example, if you have a human with antlers: Only mask the humans head and put "antlers" in the negative prompt. Fix all the errors step by step... this can take some time! Creating nice duos is hard work!

So... Does it work like this for example? (Human, white skin, blonde hair, BREAK, wolf, female)?

masterbeaster50 said:
So... Does it work like this for example? (Human, white skin, blonde hair, BREAK, wolf, female)?

Yes, but I also suggest an extra block for general tags. Of course you would want to use more tags for every character... and put the tags, that are valid for both characters in the general block and also tags that describe the quality/style of the image and the background.

But like I said: You won't get good results with this "out of the box". Most likely you will have to run a large batch to get something that is somewhat similar to what you imagined. And when you run a large batch, you waste lots of electricity. So the better alternative is, to take one of the "failed" generations and inpaint it. If your female wolf happens to have long blonde hair, just mask the head/hair and put "blonde hair" in the neg. prompt. And fix everything that is wrong step by step.

By the way: Joining the furry diffusion server might be a good idea. Topics like this are discussed there almost daily. You can ask for help or just read what other people asked in the prompt-workshop and learn a lot! You can also find me there. Is probably easier to chat there, than in this forum.

silvicultor said:
Yes, but I also suggest an extra block for general tags. Of course you would want to use more tags for every character... and put the tags, that are valid for both characters in the general block and also tags that describe the quality/style of the image and the background.

But like I said: You won't get good results with this "out of the box". Most likely you will have to run a large batch to get something that is somewhat similar to what you imagined. And when you run a large batch, you waste lots of electricity. So the better alternative is, to take one of the "failed" generations and inpaint it. If your female wolf happens to have long blonde hair, just mask the head/hair and put "blonde hair" in the neg. prompt. And fix everything that is wrong step by step.

By the way: Joining the furry diffusion server might be a good idea. Topics like this are discussed there almost daily. You can ask for help or just read what other people asked in the prompt-workshop and learn a lot! You can also find me there. Is probably easier to chat there, than in this forum.

It worked most of the time. But now I still have problems when it comes to the points of views. I end up getting male POVs, despite wanting to show both my human male and anthro female's faces when they're having sex or when my female anthro is giving my human male a tit job for example

I have had luck splitting the subjects up, it of course depends on which checkpoint you use but this is what has worked for me:

Subject 1, white fur, species, gender,
BREAK
Subject 2, human, gender
BREAK
Setting, bedroom, daytime

masterbeaster50 said:
It worked most of the time. But now I still have problems when it comes to the points of views. I end up getting male POVs, despite wanting to show both my human male and anthro female's faces when they're having sex or when my female anthro is giving my human male a tit job for example

There is no easy solution for this problem. You could use controlnet to influence pose/POV. But that is complicated.
There are two other possible approaches:

1. (like I already suggested) Make a sketch and feed it to img2img. You don’t have to be a real artist for that. The difficult part is to find the right denoising strength.

2. Using the “BREAK-method”. You already know how it works. If your batch is finished you have to pick one good image to inpaint, take one with the right pose/POV, even if the characters look weird. Because it’s much easier to fix a broken body with inpainting than changing the whole pose.