File size: 4,468 Bytes
960617b
 
 
 
 
 
 
 
 
 
 
0c9b27d
 
 
 
960617b
 
 
0c9b27d
960617b
0c9b27d
960617b
0c9b27d
960617b
0c9b27d
960617b
0c9b27d
960617b
4460f9b
e951fec
 
 
 
 
 
4460f9b
 
 
 
 
 
 
 
e951fec
960617b
 
 
 
cd2e780
 
 
 
562e3f6
960617b
22feceb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
960617b
 
 
 
0c9b27d
 
 
960617b
 
 
36ecf12
 
 
 
 
0c9b27d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
---
base_model: stabilityai/stable-diffusion-xl-base-1.0
library_name: diffusers
tags:
- stable-diffusion-xl
- stable-diffusion-xl-diffusers
- text-to-image
- diffusers
- diffusers-training
- lora
inference: true
datasets:
- AdamLucek/oldbookillustrations-small
language:
- en
---


# 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](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0). The weights were fine-tuned on the [AdamLucek/oldbookillustrations-small dataset](https://huggingface.co/datasets/AdamLucek/oldbookillustrations-small).

LoRA for the text encoder was enabled: **True**.

Special VAE used for training: [madebyollin/sdxl-vae-fp16-fix](https://huggingface.co/madebyollin/sdxl-vae-fp16-fix).

## Example Images

<!-- *"An Apple"*
<img src="https://cdn-uploads.huggingface.co/production/uploads/65ba68a15d2ef0a4b2c892b4/w82UDqXeWNK0WRPpCk0Wk.png" width=300>
*"A Flower Wreath"*
<img src="https://cdn-uploads.huggingface.co/production/uploads/65ba68a15d2ef0a4b2c892b4/PT9tb-2WDz9eb7Rv8vMsu.png" width=300>
*"A view down an alley in New York"*
<img src="https://cdn-uploads.huggingface.co/production/uploads/65ba68a15d2ef0a4b2c892b4/Sr9cHjKpXFRPFH4ODW6Sz.png" width=300>
*"An office setting with a desk and papers on it, with a view out the window above the desk into the town"*
<img src="https://cdn-uploads.huggingface.co/production/uploads/65ba68a15d2ef0a4b2c892b4/A8SETPQoBMs2oay7X6iwp.png" width=300> -->

<img src="https://cdn-uploads.huggingface.co/production/uploads/65ba68a15d2ef0a4b2c892b4/DIxEBANVrf1JSGdm_ldx-.jpeg" width=500>

*"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

<div style="display: flex; align-items: center;">
  <img src="https://upload.wikimedia.org/wikipedia/commons/d/d0/Google_Colaboratory_SVG_Logo.svg" width="100">
  <a href="https://colab.research.google.com/drive/1SMLhqfW2xjOK8Ep3R2FGT-xUyX50DgZh?usp=sharing" style="margin-left: 10px;">COLAB Notebook Here</a>
</div>

```python
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**
<a href="https://youtu.be/v89kB4OScOA">
  <img src="https://i.imgur.com/fW6hHu2.png" width="350">
</a>

Trained on a single a100 using the [Diffuser's package](https://github.com/huggingface/diffusers/blob/main/examples/text_to_image/train_text_to_image_lora_sdxl.py), documentation available [here](https://huggingface.co/docs/diffusers/main/en/training/lora).

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"
```