import torch
from diffusers import DiffusionPipeline
# switch to "mps" for apple devices
pipe = DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0", dtype=torch.bfloat16, device_map="cuda")
pipe.load_lora_weights("ruggsea/roman-mosaics-lora")
prompt = "bird mosaic"
image = pipe(prompt).images[0]SDXL LoRA DreamBooth - ruggsea/lora-trained-xl
- Prompt
- bird mosaic
Model description
These are the LoRA adaptation weights for stabilityai/stable-diffusion-xl-base-1.0, trained by me on a dataset of pictures of roman mosaics in order to capture their particular style.
The weights were trained using the SDXL Lora Dreambooth training script from the Diffusers library. The training parameters and other details can be found in this repo.
LoRA for the text encoder was not enabled.
A special VAE, madebyollin/sdxl-vae-fp16-fix, was used for training as specified by the Diffusers training script for Stable Diffusion XL models.
Trigger words
To generate images, use a photo of a Roman mosaic as the trigger.
Download model
The weights for this model are available in Safetensors format. You can download them from the Files & versions tab. The github repo also contains a version of the Lora converted for use with the Automatic1111 webui. You can download it by clicking here.
- Downloads last month
- 12
Model tree for ruggsea/roman-mosaics-lora
Base model
stabilityai/stable-diffusion-xl-base-1.0