---
tags:
- stable-diffusion-xl
- stable-diffusion-xl-diffusers
- text-to-image
- diffusers
- lora
- template:sd-lora
widget:

        - text: 'Orthodox church in the style of African buildings of the 6th century'
          output:
            url:
                "image_0.png"
        
        - text: 'Orthodox church in the style of African buildings of the 6th century'
          output:
            url:
                "image_1.png"
        
        - text: 'Orthodox church in the style of African buildings of the 6th century'
          output:
            url:
                "image_2.png"
        
        - text: 'Orthodox church in the style of African buildings of the 6th century'
          output:
            url:
                "image_3.png"
        
base_model: stabilityai/stable-diffusion-xl-base-1.0
instance_prompt: Orthodox church
license: openrail++
---
    
# SDXL LoRA DreamBooth - litvan/SDXL_finetuned_for_russian_churches

<Gallery />

## Model description

These are litvan/SDXL_finetuned_for_russian_churches LoRA adaption weights for stabilityai/stable-diffusion-xl-base-1.0.

The main purpose of the model: Generate Orthodox churches in different cultural and architectural codes of countries

The weights were trained  using [DreamBooth](https://dreambooth.github.io/).

LoRA for the text encoder was enabled: False.

Special VAE used for training: madebyollin/sdxl-vae-fp16-fix.

Dataset for finetuning: litvan/russian_churches_with_blip_captioning

For training were used: 3 GPU A100(80Gb)

## Trigger words

You should use Orthodox church to trigger the image generation.

## Download model

You can do this using the following lines of code:

```
from diffusers import DiffusionPipeline

pipeline = DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0").cuda()
pipeline.load_lora_weights("litvan/SDXL_finetuned_for_russian_churches")
```

### For using refiner

```
refiner = DiffusionPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-refiner-1.0",
    text_encoder_2=pipeline.text_encoder_2,
    vae=pipeline.vae,
    torch_dtype=torch.float32,
    use_safetensors=True,
).cuda()
```