Spaces:
Sleeping
Sleeping
Commit
•
4120479
1
Parent(s):
2905c29
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,109 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
from time import sleep
|
3 |
+
from diffusers import DiffusionPipeline
|
4 |
+
import torch
|
5 |
+
import json
|
6 |
+
import random
|
7 |
+
|
8 |
+
lora_list = hf_hub_download(repo_id="multimodalart/LoraTheExplorer", filename="sdxl_loras.json", repo_type="space")
|
9 |
+
|
10 |
+
with open(lora_list, "r") as file:
|
11 |
+
data = json.load(file)
|
12 |
+
sdxl_loras = [
|
13 |
+
{
|
14 |
+
"image": item["image"],
|
15 |
+
"title": item["title"],
|
16 |
+
"repo": item["repo"],
|
17 |
+
"trigger_word": item["trigger_word"],
|
18 |
+
"weights": item["weights"],
|
19 |
+
"is_compatible": item["is_compatible"],
|
20 |
+
"is_pivotal": item.get("is_pivotal", False),
|
21 |
+
"text_embedding_weights": item.get("text_embedding_weights", None),
|
22 |
+
"is_nc": item.get("is_nc", False)
|
23 |
+
}
|
24 |
+
for item in data
|
25 |
+
]
|
26 |
+
|
27 |
+
saved_names = [
|
28 |
+
hf_hub_download(item["repo"], item["weights"]) for item in sdxl_loras
|
29 |
+
]
|
30 |
+
|
31 |
+
css = '''
|
32 |
+
#title{text-align:center}
|
33 |
+
#plus_column{align-self: center}
|
34 |
+
#plus_button{font-size: 250%; text-align: center}
|
35 |
+
.gradio-container{width: 700px !important; margin: 0 auto !important}
|
36 |
+
#prompt input{width: calc(100% - 160px);border-top-right-radius: 0px;border-bottom-right-radius: 0px;}
|
37 |
+
#run_button{position:absolute;margin-top: 57px;right: 0;margin-right: 0.8em;border-bottom-left-radius: 0px;
|
38 |
+
border-top-left-radius: 0px;}
|
39 |
+
'''
|
40 |
+
|
41 |
+
pipe = DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16)
|
42 |
+
original_pipe = copy.deepcopy(pipe)
|
43 |
+
def merge_and_run(prompt, negative_prompt, shuffled_items, lora_1_scale=0.5, lora_2_scale=0.5, progress=gr.Progress(track_tqdm=True)):
|
44 |
+
pipe = copy.deepcopy(original_pipe)
|
45 |
+
pipe.load_lora_weights(shuffled_items[0]['repo'], weight_name=shuffled_items[0]['weights'])
|
46 |
+
pipe.fuse_lora(lora_1_scale)
|
47 |
+
pipe.load_lora_weights(shuffled_items[1]['repo'], weight_name=shuffled_items[1]['weights'])
|
48 |
+
pipe.fuse_lora(lora_2_scale)
|
49 |
+
|
50 |
+
pipe.to(torch_dtype=torch.float16)
|
51 |
+
pipe.to("cuda")
|
52 |
+
if negative_prompt == "":
|
53 |
+
negative_prompt = False
|
54 |
+
image = pipe(prompt=prompt, negative_prompt=negative_prompt, num_inference_steps=25, guidance_scale=7).images[0]
|
55 |
+
return image
|
56 |
+
|
57 |
+
def get_description(item):
|
58 |
+
trigger_word = item["trigger_word"]
|
59 |
+
return f"LoRA trigger word: `{trigger_word}`" if trigger_word else "LoRA trigger word: `none`, will be applied automatically", trigger_word
|
60 |
+
|
61 |
+
def shuffle_images():
|
62 |
+
compatible_items = [item for item in data if item['is_compatible']]
|
63 |
+
random.shuffle(compatible_items)
|
64 |
+
two_shuffled_items = compatible_items[:2]
|
65 |
+
title_1 = gr.update(label=two_shuffled_items[0]['title'], value=two_shuffled_items[0]['image'])
|
66 |
+
title_2 = gr.update(label=two_shuffled_items[1]['title'], value=two_shuffled_items[1]['image'])
|
67 |
+
|
68 |
+
description_1, trigger_word_1 = get_description(two_shuffled_items[0])
|
69 |
+
description_2, trigger_word_2 = get_description(two_shuffled_items[1])
|
70 |
+
|
71 |
+
prompt = gr.update(value=f"{trigger_word_1} {trigger_word_2}")
|
72 |
+
return title_1,description_1,title_2,description_2,prompt, two_shuffled_items
|
73 |
+
|
74 |
+
with gr.Blocks(css=css) as demo:
|
75 |
+
shuffled_items = gr.State()
|
76 |
+
title = gr.HTML(
|
77 |
+
'''<h1>LoRA Roulette 🎲</h1>
|
78 |
+
<h4>Two LoRAs are loaded to SDXL at random, find a way to combine them for your art 🎨</h4>
|
79 |
+
''',
|
80 |
+
elem_id="title"
|
81 |
+
)
|
82 |
+
with gr.Row():
|
83 |
+
with gr.Column(min_width=10, scale=6):
|
84 |
+
lora_1 = gr.Image(interactive=False, height=350)
|
85 |
+
lora_1_prompt = gr.Markdown()
|
86 |
+
with gr.Column(min_width=10, scale=1, elem_id="plus_column"):
|
87 |
+
plus = gr.HTML("+", elem_id="plus_button")
|
88 |
+
with gr.Column(min_width=10, scale=6):
|
89 |
+
lora_2 = gr.Image(interactive=False, height=350)
|
90 |
+
lora_2_prompt = gr.Markdown()
|
91 |
+
with gr.Row():
|
92 |
+
prompt = gr.Textbox(label="Your prompt", info="arrange the trigger words of the two LoRAs in a coherent sentence", interactive=True, elem_id="prompt")
|
93 |
+
run_btn = gr.Button("Run", elem_id="run_button")
|
94 |
+
|
95 |
+
output_image = gr.Image()
|
96 |
+
with gr.Accordion("Advanced settings", open=False):
|
97 |
+
negative_prompt = gr.Textbox(label="Negative prompt")
|
98 |
+
with gr.Row():
|
99 |
+
lora_1_scale = gr.Slider(label="LoRA 1 Scale", minimum=0, maximum=1, step=0.1, value=0.7)
|
100 |
+
lora_2_scale = gr.Slider(label="LoRa 2 Scale", minimum=0, maximum=1, step=0.1, value=0.7)
|
101 |
+
shuffle_button = gr.Button("Reshuffle LoRAs!")
|
102 |
+
|
103 |
+
demo.load(shuffle_images, inputs=[], outputs=[lora_1,lora_1_prompt,lora_2,lora_2_prompt, prompt, shuffled_items], queue=False, show_progress="hidden")
|
104 |
+
shuffle_button.click(shuffle_images, outputs=[lora_1,lora_1_prompt,lora_2,lora_2_prompt, prompt, shuffled_items], queue=False, show_progress="hidden")
|
105 |
+
|
106 |
+
run_btn.click(merge_and_run, inputs=[prompt, negative_prompt, shuffled_items, lora_1_scale, lora_2_scale], outputs=[output_image])
|
107 |
+
prompt.submit(merge_and_run, inputs=[prompt, negative_prompt, shuffled_items, lora_1_scale, lora_2_scale], outputs=[output_image])
|
108 |
+
demo.queue()
|
109 |
+
demo.launch()
|