Note: The sample images are not representative of the specific style, as these models should adapt to various styles based on the checkpoint and other Loras used. The goal of these models is to capture the various male body types.

Male Mix Pony Edition

For those who don't know, Pony already does a decent job capturing most male body types, so in some cases a lora like this isn't even necessary. This began as an SDXL lora, as SDXL doesn't seem to have as much training on male bodies, and was just converted to Pony as a test to add a little more variety to what the base model already had. The main issue with Pony is that it seems like there was a very limited "photo/realistic" image set in the training data, so it does tend to give a similar face to most males done in a realistic style from my testing.

The first two versions aren't much different, I tried to add more variety with version 2, and got some stylistic changes that I thought were interesting. Version 3 is a merge of 1 and 2, with some extra training mixed in, so it has a much different style, but doesn't do photo prompts as well as the other two.

Note for v3: I've seen some people having issues with v3, and this is likely because a bad merge was used when I made it. I have since stopped using that merge and deleted it, so I cannot say what other Loras may have been in it, and could give you errors.

Version 4 uses a completely new, and larger data set, that also includes photo/realistic images. Because of this I feel like I was able to capture the different body types better (finally some differentiation between femboy and androgynous), and also realistic images look much better (though it still leans toward whatever realistic male face the base pony model was trained on, it does try to break away from that more).

This example used the base Pony V6 model.

v1

v2

v3

v4 - This model used a more diverse data set, so it does a bit better with differentiating the body types, especially femboy and androgynous.

I also got good results from the (NSFW) GrLE mix checkpoint, which has a more consistent style, https://civitai.com/models/343190/grle-mix-ponyxl

The femboy and androgynous tags will get you similar results, but using one instead of the other can at least give you a different output. The same goes for the toned and muscular tags, and in some cases the skinny tag. The twink tag has the least amount of overlap. It's best to try different tags to find the best results, and of course better prompting gives better variety.

Note For v4: This model does an overall better job with the body types. 'Femboy' has the typical feminine body and face, while 'Androgynous' has a feminine face but a slightly more masculine body, and Twink has a more masculine, young face, with a body type that usually goes between skinny and toned. 'Skinny' does a better job of keeping to thinner bodies, 'Toned' is semi-muscular, while 'Muscular' is closer to the bara body type.

As with the SDXL version I didn't include any realistic images or photos, so without using a checkpoint, or additional lora specifically made for photos, you likely will only get the digital art style.

For the images used, it was a mix of different styles meant to capture the body types and a general digital art aesthetic, plus whatever style the Pony base model added into it. So I'll just call this "Pony style".

This example uses the One For All pony checkpoint, https://civitai.com/models/349062/one-for-all-pony-fantasy-dpovae though I also recommend using the Halcyon, https://civitai.com/models/668260/halcyon-pony or Zonkey, https://civitai.com/models/405136/zonkey-realism for realistic images with any of these loras.

v1

v2

v3

v4 - As stated, I did add photo/realistic images to the dataset, so this model does much better with realism.

I hope others can find some use for this model.

All of my models may be used for other training and merges.