AdamLucek/sdxl-base-1.0-oldbookillustrations-lora

LoRA Weights on Old Book Illustrations for Stable Diffusion XL Base 1.0

These are LoRA adaption weights for stabilityai/stable-diffusion-xl-base-1.0. The weights were fine-tuned on the AdamLucek/oldbookillustrations-small dataset.

LoRA for the text encoder was enabled: True.

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

Example Images

"An Apple"
"A Flower Wreath"
"A view down an alley in New York"
"An office setting with a desk and papers on it, with a view out the window above the desk into the town"

Intended uses & limitations

How to use

COLAB Notebook Here

from diffusers import DiffusionPipeline
import torch

# Load Stable Diffusion XL Base1.0
pipe = DiffusionPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0",
    torch_dtype=torch.float16,
    variant="fp16",
    use_safetensors=True
).to("cuda")

# Optional CPU offloading to save some GPU Memory
pipe.enable_model_cpu_offload()

# Loading Trained Old Book Illustrations LoRA Weights
pipe.load_lora_weights("AdamLucek/sdxl-base-1.0-oldbookillustrations-lora")

# Generate an Image
prompt = "An Apple"

image = pipe(
    prompt = prompt,
    num_inference_steps=50,
    height=1024,
    width=1024,
).images[0]

# Save the image
image.save("SDXL_OldBookIllustrations.png")

Limitations and bias

Note: See original Stable Diffusion XL Base 1.0 page for additional limitations and biases

Note: First try with tuning hyperparameters

Training details

Video Overview

Trained on a single a100 using the Diffuser's package, documentation available here.

Training script used:

accelerate launch train_text_to_image_lora_sdxl.py \
  --pretrained_model_name_or_path=stabilityai/stable-diffusion-xl-base-1.0 \
  --pretrained_vae_model_name_or_path="madebyollin/sdxl-vae-fp16-fix" \
  --dataset_name="AdamLucek/oldbookillustrations-small" \
  --validation_prompt="An inventor tinkers with a complex machine in his workshop, oblivious to the setting sun outside" \
  --num_validation_images=4 \
  --validation_epochs=1 \
  --output_dir="output/sdxl-base-1.0-oldbookillustrations-lora" \
  --resolution=1024 \
  --center_crop \
  --random_flip \
  --train_text_encoder \
  --train_batch_size=1 \
  --num_train_epochs=10 \
  --checkpointing_steps=500 \
  --gradient_accumulation_steps=4 \
  --learning_rate=1e-04 \
  --lr_warmup_steps=0 \
  --report_to="wandb" \
  --dataloader_num_workers=8 \
  --allow_tf32 \
  --mixed_precision="fp16" \
  --push_to_hub \
  --hub_model_id="sdxl-base-1.0-oldbookillustrations-lora"