Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -1,103 +1,143 @@
|
|
1 |
import spaces
|
|
|
2 |
import time
|
3 |
import torch
|
4 |
import gradio as gr
|
5 |
from PIL import Image
|
6 |
-
from huggingface_hub import hf_hub_download
|
7 |
from src_inference.pipeline import FluxPipeline
|
8 |
from src_inference.lora_helper import set_single_lora
|
9 |
-
import random
|
10 |
|
11 |
-
|
12 |
-
|
13 |
-
|
14 |
-
|
15 |
-
|
16 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
17 |
|
18 |
-
# Initialize the pipeline with the model
|
19 |
-
pipe = FluxPipeline.from_pretrained(base_path, torch_dtype=torch.bfloat16).to("cuda")
|
20 |
-
|
21 |
-
# Set LoRA weights
|
22 |
-
set_single_lora(pipe.transformer, omni_consistency_path, lora_weights=[1], cond_size=512)
|
23 |
-
|
24 |
-
# Function to clear cache
|
25 |
-
def clear_cache(transformer):
|
26 |
-
for name, attn_processor in transformer.attn_processors.items():
|
27 |
-
attn_processor.bank_kv.clear()
|
28 |
-
|
29 |
-
# Function to download all LoRAs in advance
|
30 |
def download_all_loras():
|
31 |
lora_names = [
|
32 |
-
"3D_Chibi", "American_Cartoon", "Chinese_Ink",
|
33 |
-
"
|
34 |
-
"
|
35 |
-
"
|
36 |
-
"
|
37 |
-
"Rick_Morty", "Snoopy", "Van_Gogh", "Vector"
|
38 |
]
|
39 |
-
for
|
40 |
-
hf_hub_download(
|
41 |
-
|
42 |
-
|
43 |
-
|
44 |
-
|
45 |
download_all_loras()
|
46 |
|
47 |
-
|
|
|
|
|
|
|
48 |
@spaces.GPU()
|
49 |
-
def generate_image(
|
50 |
-
|
51 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
52 |
|
53 |
-
# Load the specific LoRA weights
|
54 |
pipe.unload_lora_weights()
|
55 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
56 |
|
57 |
-
|
58 |
-
spatial_image = [uploaded_image.convert("RGB")]
|
59 |
subject_images = []
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
# Generate the image
|
64 |
-
image = pipe(
|
65 |
prompt,
|
66 |
-
height=(
|
67 |
-
width=(
|
68 |
guidance_scale=guidance_scale,
|
69 |
num_inference_steps=num_inference_steps,
|
70 |
max_sequence_length=512,
|
71 |
-
generator=
|
72 |
spatial_images=spatial_image,
|
73 |
subject_images=subject_images,
|
74 |
cond_size=512,
|
75 |
).images[0]
|
|
|
76 |
|
77 |
-
end_time = time.time()
|
78 |
-
elapsed_time = end_time - start_time
|
79 |
-
print(f"code running time: {elapsed_time} s")
|
80 |
-
|
81 |
-
# Clear cache after generation
|
82 |
clear_cache(pipe.transformer)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
83 |
|
84 |
-
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
|
95 |
-
|
96 |
-
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
header = """
|
101 |
<div style="text-align: center; display: flex; justify-content: left; gap: 5px;">
|
102 |
<a href="https://arxiv.org/abs/2505.18445"><img src="https://img.shields.io/badge/ariXv-2505.18445-A42C25.svg" alt="arXiv"></a>
|
103 |
<a href="https://huggingface.co/showlab/OmniConsistency"><img src="https://img.shields.io/badge/🤗_HuggingFace-Model-ffbd45.svg" alt="HuggingFace"></a>
|
@@ -105,58 +145,55 @@ header = """
|
|
105 |
</div>
|
106 |
"""
|
107 |
|
108 |
-
# Gradio interface setup
|
109 |
-
def create_gradio_interface():
|
110 |
-
lora_names = [
|
111 |
-
"3D_Chibi", "American_Cartoon", "Chinese_Ink",
|
112 |
-
"Clay_Toy", "Fabric", "Ghibli", "Irasutoya",
|
113 |
-
"Jojo", "LEGO", "Line", "Macaron",
|
114 |
-
"Oil_Painting", "Origami", "Paper_Cutting",
|
115 |
-
"Picasso", "Pixel", "Poly", "Pop_Art",
|
116 |
-
"Rick_Morty", "Snoopy", "Van_Gogh", "Vector"
|
117 |
-
]
|
118 |
-
|
119 |
with gr.Blocks() as demo:
|
120 |
gr.Markdown("# OmniConsistency LoRA Image Generation")
|
121 |
-
gr.Markdown("Select a LoRA, enter a prompt, and upload an image to generate a new image with OmniConsistency.
|
122 |
gr.HTML(header)
|
|
|
123 |
with gr.Row():
|
124 |
with gr.Column(scale=1):
|
125 |
-
lora_dropdown = gr.Dropdown(
|
126 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
127 |
image_input = gr.Image(type="pil", label="Upload Image")
|
128 |
with gr.Column(scale=1):
|
129 |
output_image = gr.ImageSlider(label="Generated Image")
|
130 |
-
|
131 |
-
|
132 |
-
guidance_slider = gr.Slider(
|
133 |
-
|
134 |
-
|
135 |
-
|
136 |
-
|
137 |
-
|
|
|
|
|
138 |
gr.Examples(
|
139 |
examples=examples,
|
140 |
-
inputs=[lora_dropdown, prompt_box, image_input,
|
|
|
141 |
outputs=output_image,
|
142 |
fn=generate_image,
|
143 |
cache_examples=False,
|
144 |
label="Examples"
|
145 |
)
|
146 |
|
147 |
-
|
148 |
fn=generate_image,
|
149 |
-
inputs=[
|
150 |
-
|
151 |
-
width_box, height_box, guidance_slider,
|
152 |
-
steps_slider, seed_slider
|
153 |
-
],
|
154 |
outputs=output_image
|
155 |
)
|
156 |
-
|
157 |
return demo
|
158 |
|
159 |
-
|
160 |
-
|
161 |
-
|
162 |
-
interface.launch()
|
|
|
1 |
import spaces
|
2 |
+
import os
|
3 |
import time
|
4 |
import torch
|
5 |
import gradio as gr
|
6 |
from PIL import Image
|
7 |
+
from huggingface_hub import hf_hub_download, list_repo_files
|
8 |
from src_inference.pipeline import FluxPipeline
|
9 |
from src_inference.lora_helper import set_single_lora
|
|
|
10 |
|
11 |
+
BASE_PATH = "black-forest-labs/FLUX.1-dev"
|
12 |
+
LOCAL_LORA_DIR = "./LoRAs"
|
13 |
+
CUSTOM_LORA_DIR = "./Custom_LoRAs"
|
14 |
+
os.makedirs(LOCAL_LORA_DIR, exist_ok=True)
|
15 |
+
os.makedirs(CUSTOM_LORA_DIR, exist_ok=True)
|
16 |
+
|
17 |
+
print("downloading OmniConsistency base LoRA …")
|
18 |
+
omni_consistency_path = hf_hub_download(
|
19 |
+
repo_id="showlab/OmniConsistency",
|
20 |
+
filename="OmniConsistency.safetensors",
|
21 |
+
local_dir="./Model"
|
22 |
+
)
|
23 |
+
|
24 |
+
print("loading base pipeline …")
|
25 |
+
pipe = FluxPipeline.from_pretrained(
|
26 |
+
BASE_PATH, torch_dtype=torch.bfloat16
|
27 |
+
).to("cuda")
|
28 |
+
set_single_lora(pipe.transformer, omni_consistency_path,
|
29 |
+
lora_weights=[1], cond_size=512)
|
30 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
31 |
def download_all_loras():
|
32 |
lora_names = [
|
33 |
+
"3D_Chibi", "American_Cartoon", "Chinese_Ink", "Clay_Toy",
|
34 |
+
"Fabric", "Ghibli", "Irasutoya", "Jojo", "LEGO", "Line",
|
35 |
+
"Macaron", "Oil_Painting", "Origami", "Paper_Cutting",
|
36 |
+
"Picasso", "Pixel", "Poly", "Pop_Art", "Rick_Morty",
|
37 |
+
"Snoopy", "Van_Gogh", "Vector"
|
|
|
38 |
]
|
39 |
+
for name in lora_names:
|
40 |
+
hf_hub_download(
|
41 |
+
repo_id="showlab/OmniConsistency",
|
42 |
+
filename=f"LoRAs/{name}_rank128_bf16.safetensors",
|
43 |
+
local_dir=LOCAL_LORA_DIR,
|
44 |
+
)
|
45 |
download_all_loras()
|
46 |
|
47 |
+
def clear_cache(transformer):
|
48 |
+
for _, attn_processor in transformer.attn_processors.items():
|
49 |
+
attn_processor.bank_kv.clear()
|
50 |
+
|
51 |
@spaces.GPU()
|
52 |
+
def generate_image(
|
53 |
+
lora_name,
|
54 |
+
custom_repo_id,
|
55 |
+
prompt,
|
56 |
+
uploaded_image,
|
57 |
+
width, height,
|
58 |
+
guidance_scale,
|
59 |
+
num_inference_steps,
|
60 |
+
seed
|
61 |
+
):
|
62 |
+
width, height = int(width), int(height)
|
63 |
+
generator = torch.Generator("cpu").manual_seed(seed)
|
64 |
+
|
65 |
+
if custom_repo_id and custom_repo_id.strip():
|
66 |
+
repo_id = custom_repo_id.strip()
|
67 |
+
try:
|
68 |
+
files = list_repo_files(repo_id)
|
69 |
+
print("using custom LoRA from:", repo_id)
|
70 |
+
safetensors_files = [f for f in files if f.endswith(".safetensors")]
|
71 |
+
print("found safetensors files:", safetensors_files)
|
72 |
+
if not safetensors_files:
|
73 |
+
raise ValueError("No .safetensors files were found in this repo")
|
74 |
+
fname = safetensors_files[0]
|
75 |
+
lora_path = hf_hub_download(
|
76 |
+
repo_id=repo_id,
|
77 |
+
filename=fname,
|
78 |
+
local_dir=CUSTOM_LORA_DIR,
|
79 |
+
)
|
80 |
+
except Exception as e:
|
81 |
+
raise gr.Error(f"Load custom LoRA failed: {e}")
|
82 |
+
else:
|
83 |
+
lora_path = os.path.join(
|
84 |
+
f"{LOCAL_LORA_DIR}/LoRAs", f"{lora_name}_rank128_bf16.safetensors"
|
85 |
+
)
|
86 |
|
|
|
87 |
pipe.unload_lora_weights()
|
88 |
+
try:
|
89 |
+
pipe.load_lora_weights(
|
90 |
+
os.path.dirname(lora_path),
|
91 |
+
weight_name=os.path.basename(lora_path)
|
92 |
+
)
|
93 |
+
except Exception as e:
|
94 |
+
raise gr.Error(f"Load LoRA failed: {e}")
|
95 |
|
96 |
+
spatial_image = [uploaded_image.convert("RGB")]
|
|
|
97 |
subject_images = []
|
98 |
+
start = time.time()
|
99 |
+
out_img = pipe(
|
|
|
|
|
|
|
100 |
prompt,
|
101 |
+
height=(height // 8) * 8,
|
102 |
+
width=(width // 8) * 8,
|
103 |
guidance_scale=guidance_scale,
|
104 |
num_inference_steps=num_inference_steps,
|
105 |
max_sequence_length=512,
|
106 |
+
generator=generator,
|
107 |
spatial_images=spatial_image,
|
108 |
subject_images=subject_images,
|
109 |
cond_size=512,
|
110 |
).images[0]
|
111 |
+
print(f"inference time: {time.time()-start:.2f}s")
|
112 |
|
|
|
|
|
|
|
|
|
|
|
113 |
clear_cache(pipe.transformer)
|
114 |
+
return uploaded_image, out_img
|
115 |
+
|
116 |
+
# =============== Gradio UI ===============
|
117 |
+
def create_interface():
|
118 |
+
demo_lora_names = [
|
119 |
+
"3D_Chibi", "American_Cartoon", "Chinese_Ink", "Clay_Toy",
|
120 |
+
"Fabric", "Ghibli", "Irasutoya", "Jojo", "LEGO", "Line",
|
121 |
+
"Macaron", "Oil_Painting", "Origami", "Paper_Cutting",
|
122 |
+
"Picasso", "Pixel", "Poly", "Pop_Art", "Rick_Morty",
|
123 |
+
"Snoopy", "Van_Gogh", "Vector"
|
124 |
+
]
|
125 |
|
126 |
+
# Example data
|
127 |
+
examples = [
|
128 |
+
["3D_Chibi", "", "3D Chibi style, Two smiling colleagues enthusiastically high-five in front of a whiteboard filled with technical notes about multimodal learning, reflecting a moment of success and collaboration at OpenAI.",
|
129 |
+
Image.open("./test_imgs/00.png"), 680, 1024, 3.5, 24, 42],
|
130 |
+
["Clay_Toy", "", "Clay Toy style, Three team members from OpenAI are gathered around a laptop in a cozy, festive setting, with holiday decorations in the background; one waves cheerfully while the others engage in light conversation, reflecting a relaxed and collaborative atmosphere.",
|
131 |
+
Image.open("./test_imgs/01.png"), 560, 1024, 3.5, 24, 42],
|
132 |
+
["American_Cartoon", "", "American Cartoon style, In a dramatic and comedic moment from a classic Chinese film, an intense elder with a white beard and red hat grips a younger man, declaring something with fervor, while the subtitle at the bottom reads 'I want them all' — capturing both tension and humor.",
|
133 |
+
Image.open("./test_imgs/02.png"), 568, 1024, 3.5, 24, 42],
|
134 |
+
["Origami", "", "Origami style, A thrilled fan wearing a Portugal football kit poses energetically with a smiling Cristiano Ronaldo, who gives a thumbs-up, as they stand side by side in a casual, cheerful moment—capturing the excitement of meeting a football legend.",
|
135 |
+
Image.open("./test_imgs/03.png"), 768, 672, 3.5, 24, 42],
|
136 |
+
["Vector", "", "Vector style, A man glances admiringly at a passing woman, while his girlfriend looks at him in disbelief, perfectly capturing the theme of shifting attention and misplaced priorities in a humorous, relatable way.",
|
137 |
+
Image.open("./test_imgs/04.png"), 512, 1024, 3.5, 24, 42]
|
138 |
+
]
|
139 |
+
|
140 |
+
header = """
|
|
|
|
|
141 |
<div style="text-align: center; display: flex; justify-content: left; gap: 5px;">
|
142 |
<a href="https://arxiv.org/abs/2505.18445"><img src="https://img.shields.io/badge/ariXv-2505.18445-A42C25.svg" alt="arXiv"></a>
|
143 |
<a href="https://huggingface.co/showlab/OmniConsistency"><img src="https://img.shields.io/badge/🤗_HuggingFace-Model-ffbd45.svg" alt="HuggingFace"></a>
|
|
|
145 |
</div>
|
146 |
"""
|
147 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
148 |
with gr.Blocks() as demo:
|
149 |
gr.Markdown("# OmniConsistency LoRA Image Generation")
|
150 |
+
gr.Markdown("Select a LoRA, enter a prompt, and upload an image to generate a new image with OmniConsistency.")
|
151 |
gr.HTML(header)
|
152 |
+
|
153 |
with gr.Row():
|
154 |
with gr.Column(scale=1):
|
155 |
+
lora_dropdown = gr.Dropdown(
|
156 |
+
demo_lora_names, label="Select built-in LoRA")
|
157 |
+
custom_repo_box = gr.Textbox(
|
158 |
+
label="Enter Custom LoRA",
|
159 |
+
placeholder="LoRA Hugging Face path (e.g., 'username/repo_name')",
|
160 |
+
info="If you want to use a custom LoRA, enter its Hugging Face repo ID here and built-in LoRA will be Overridden. Leave empty to use built-in LoRAs. [Check the list of FLUX LoRAs](https://huggingface.co/models?other=base_model:adapter:black-forest-labs/FLUX.1-dev)"
|
161 |
+
)
|
162 |
+
prompt_box = gr.Textbox(label="Prompt",
|
163 |
+
placeholder="Enter your prompt here",
|
164 |
+
info="Remember to include the necessary trigger words if you're using a custom LoRA."
|
165 |
+
)
|
166 |
image_input = gr.Image(type="pil", label="Upload Image")
|
167 |
with gr.Column(scale=1):
|
168 |
output_image = gr.ImageSlider(label="Generated Image")
|
169 |
+
height_box = gr.Textbox(value="1024", label="Height")
|
170 |
+
width_box = gr.Textbox(value="1024", label="Width")
|
171 |
+
guidance_slider = gr.Slider(
|
172 |
+
0.1, 20, value=3.5, step=0.1, label="Guidance Scale")
|
173 |
+
steps_slider = gr.Slider(
|
174 |
+
1, 50, value=25, step=1, label="Inference Steps")
|
175 |
+
seed_slider = gr.Slider(
|
176 |
+
1, 2_147_483_647, value=42, step=1, label="Seed")
|
177 |
+
gen_btn = gr.Button("Generate")
|
178 |
+
|
179 |
gr.Examples(
|
180 |
examples=examples,
|
181 |
+
inputs=[lora_dropdown, custom_repo_box, prompt_box, image_input,
|
182 |
+
height_box, width_box, guidance_slider, steps_slider, seed_slider],
|
183 |
outputs=output_image,
|
184 |
fn=generate_image,
|
185 |
cache_examples=False,
|
186 |
label="Examples"
|
187 |
)
|
188 |
|
189 |
+
gen_btn.click(
|
190 |
fn=generate_image,
|
191 |
+
inputs=[lora_dropdown, custom_repo_box, prompt_box, image_input,
|
192 |
+
width_box, height_box, guidance_slider, steps_slider, seed_slider],
|
|
|
|
|
|
|
193 |
outputs=output_image
|
194 |
)
|
|
|
195 |
return demo
|
196 |
|
197 |
+
if __name__ == "__main__":
|
198 |
+
demo = create_interface()
|
199 |
+
demo.launch()
|
|