A variant of my bosmer LoRA. This one is for now the main one, although I have just released an alternate which is perhaps better at making weirder renders. Generally medium weights work well on this one, use example images for some ideas regarding prompting. There are some issues in this version from biases in the training set, I used images with a lot of superfluous tags due to how I generated all the images. In future iterations I will use a smaller more properly labelled set and hopefully improve the accuracy of the model. Thanks for trying this one out though!

Files with SHA256: cd125068cf947c85927677da0eb38336103f32c9c287bf5e75ca0ed5ee971977