Forgot the main reason for using AI and this is just a little reminder not to take it to serious. Helped me quite a few times to determine how well some models work or how flexible they are, because offsetting this thing and all the quirks it introduces is not easy.

Also, maybe even more crucial, it's quite good at defiling other peoples LoRA's :)

V4.0

Trying to make the model a bit more compatible and see with what i can get away with. Used lower CFG Scale and more steps. I really wanted to push it a little bit and see what in can do using different prompts that have nothing to do with the main subject so i can see what it does. At i higher higher strength and without using fox as a trigger, it's a lot more flexible. You can change the expressions in the default position easier, but it keeps the overall idea of the main subject ( like the paws down and weird feet and strange look )

I arranged the preview images so that you can see the difference between the main subject and what happens if you change the prompts and the character.

I see this model as a testing base, because i wanna know what it does combined with different models, other Loras and weird prompts. I included a few pictures with the same prompts i used for V3.0 to see the difference.

All tests and settings are based on the model i used.

Lower CFG Scale and more steps seem to do wonders
excluding the word fox makes it much more flexible ( which is why i will exclude it in the next version )
changing the position ( or weight ) of the main trigger word makes a huge difference
most of the time CivitAI thinks the images include a wolf, dog or cat. Guess that has manly to do with the main subject being an abomination :D... putting does things into negative helps sometimes. A big head and a long neck seems to be a problem too ( or a wanted effect :D )
has the tendency to add unwanted stuff if the prompt does not tell it what to do ( guess that is based on the training images )

The description is pretty much a reminder for myself, so i can look up what i did and how other LoRAs i want to train can benefit from it.

V3.0

Usually i wouldn't post anything i trained because it's kinda tricky to handle and extremely rigged when it comes to prompts or which model works. Also a lot of plugins for 1111 play a big role sometimes. They work for me but maybe not for others.

Just a little test to see what i need to change in order to make easier to use and more compatible and flexible. I know that the .txt files and source pictures need a bit more curating, but like i said, it works for me.

This one here was made for a friend ( and for myself ofc ) that asked me about it. There is a model on Civitai that pretty much has the same contend, but it's SD1.5. Old meme is old, but i like the random results you get. Some LoRAs I've trained pretty much make no sense, yet they change the composition and style of an image drastically at lower value... like most LoRAs i guess.

Don't mind some of the words i've used in the example images. Most of the time they did nothing, nonetheless it steered the images into a certain direction. Just for my own entertainment and to see what happens.

I will probably use this model for further training.

-Trained on BaseXL

Download model

Weights for this model are available in Safetensors format.

Download them in the Files & versions tab.

Use it with the 🧨 diffusers library

from diffusers import AutoPipelineForText2Image
import torch

device = "cuda" if torch.cuda.is_available() else "cpu"

pipeline = AutoPipelineForText2Image.from_pretrained('black-forest-labs/FLUX.1-dev', torch_dtype=torch.bfloat16).to(device)
pipeline.load_lora_weights('dougduggarts/flux-1-d-sdxl-stoned-fox', weight_name='- Flux1 - stoned_fox_V1.0.safetensors')
image = pipeline('tapestry_style fox derpii normannorum ducem ,  embroidered and text,,     ').images[0]

For more details, including weighting, merging and fusing LoRAs, check the documentation on loading LoRAs in diffusers