Big Centaur Lora XL

Hello, friends!

Some of my subscribers have long wanted to try out my centaur models and more. Therefore, this day has come when I will share my models for real.

This model will allow you to create centaurs in quite flexible settings. I even managed to make centaurs that weren't even in the training data.

I must say right away that I am not a cool ML programmer and I have not been able to train this model to create perfect pictures. 90% of the images will have various artifacts and defects. Therefore, this model is suitable for those who are ready to spend even more time with processing after generating the image.

To generate images, I recommend use DreamShaperXL.

You can use almost any style.

Version 7

The model is now in Pony! So this model is best used with Pony models. Well, this model also works well with illustrious.

The model is even better at making various centaurs and taurs, the poses and scenes have been improved.

Version 5

Another update of the model. I trained this model sometime in October-November 202>

There are many new features in this version that will make your centaur creation experience even better!

For example:

- New centaurs (dinosaurs, dragons, deer)

- Numerous improvements and fixes

- Slightly improved poses

- Clothing for the lower body (belts, saddles and sometimes armor may appear on girls' bodies)

- Supine poses, , and so on will most likely not work on most checkpoints, but at least when working with checkpoints that can do such poses (Pony Diffusion), this Lora model may come in handy. Keywords for supine poses: lying on back.

- Improved the very essence of centaurs. The model can sometimes make centaurs that were not in the dataset (camels, sharks, whales, even bikes)

- Horseback riding appeared (it works poorly, such pictures will take a long time to refine)

For this version, I have collected a fairly large dataset (about 700 images), which consists of about 30-40% of my images.

Don't be afraid to experiment with this model, as well as try other Checkpoints, sometimes the result of this model makes incredibly unique results, believe me, I made a lot of pictures using this model.

It is extremely important to write the desired body of a girl, for example, "female centaur with brown horse body". The word "Centaur" is not enough.

Also, the pictures depend very much on the text you write. For example, if you want a close-up, then do not forget to remove the "hooves" or the elements of the environment associated with the floor, as well as add the word "close-up" itself.

This model works perfectly with Turbo-SDXL. Recommended weights for this model: 0.6-0.9

Horse model 04

Also, to improve the quality of horse , I made a separate model.

This model doesn't work very well. This model was conceived as an inpaint model, but since I'm bad at training models, it doesn't work the way I wanted it to.

In any case, this model works best with inpaint. If you use this model as a regular Lora, then you will get bad pictures.

But nevertheless, you can use this model with small weights so that the are better visible.

My profile on DeviantArt: https://www.deviantart.com/moondroch

If you want to support me, view my work in early access, and also try out my models, which I have not published anywhere except Boosty, then you can do it here: https://boosty.to/moondroch