Spaces:
Running
on
Zero
Running
on
Zero
alfredplpl
commited on
Commit
•
518d5a1
1
Parent(s):
452bb51
Update app.py
Browse files
app.py
CHANGED
@@ -0,0 +1,156 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Reference: https://huggingface.co/spaces/black-forest-labs/FLUX.1-schnell/blob/main/app.py
|
2 |
+
import spaces
|
3 |
+
import gradio as gr
|
4 |
+
import numpy as np
|
5 |
+
import random
|
6 |
+
import torch
|
7 |
+
import torch
|
8 |
+
from diffusers import Transformer2DModel, PixArtSigmaPipeline, AutoencoderKL, DPMSolverMultistepScheduler, DDIMScheduler, EulerAncestralDiscreteScheduler, DPMSolverSDEScheduler
|
9 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig, QuantoConfig, EetqConfig
|
10 |
+
|
11 |
+
device = "cuda"
|
12 |
+
weight_dtype = torch.float32
|
13 |
+
weight_dtype_te = torch.bfloat16
|
14 |
+
MAX_SEED = np.iinfo(np.int32).max
|
15 |
+
|
16 |
+
vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=weight_dtype)
|
17 |
+
scheduler=DPMSolverMultistepScheduler()
|
18 |
+
pipe = PixArtSigmaPipeline(
|
19 |
+
vae=vae,
|
20 |
+
tokenizer=None,
|
21 |
+
text_encoder=None,
|
22 |
+
transformer=transformer,
|
23 |
+
scheduler=scheduler
|
24 |
+
)
|
25 |
+
|
26 |
+
pipe.to(device)
|
27 |
+
|
28 |
+
tokenizer = AutoTokenizer.from_pretrained("cyberagent/calm2-7b")
|
29 |
+
text_encoder = AutoModelForCausalLM.from_pretrained(
|
30 |
+
"cyberagent/calm2-7b",
|
31 |
+
torch_dtype=weight_dtype_te,
|
32 |
+
device_map=device
|
33 |
+
)
|
34 |
+
|
35 |
+
@spaces.GPU()
|
36 |
+
def infer(prompt, seed=42, randomize_seed=False, width=512, height=512, num_inference_steps=20, progress=gr.Progress(track_tqdm=True)):
|
37 |
+
if randomize_seed:
|
38 |
+
seed = random.randint(0, MAX_SEED)
|
39 |
+
generator = torch.Generator().manual_seed(seed)
|
40 |
+
with torch.no_grad():
|
41 |
+
pos_ids = tokenizer(
|
42 |
+
prompt, max_length=512, padding="max_length", truncation=True, return_tensors="pt",
|
43 |
+
).to(device)
|
44 |
+
pos_emb = text_encoder(pos_ids.input_ids, output_hidden_states=True, attention_mask=pos_ids.attention_mask)
|
45 |
+
pos_emb = pos_emb.hidden_states[-1]
|
46 |
+
neg_ids = tokenizer(
|
47 |
+
"", max_length=512, padding="max_length", truncation=True, return_tensors="pt",
|
48 |
+
).to(device)
|
49 |
+
neg_emb = text_encoder(neg_ids.input_ids, output_hidden_states=True, attention_mask=neg_ids.attention_mask)
|
50 |
+
neg_emb = neg_emb.hidden_states[-1]
|
51 |
+
|
52 |
+
image = pipe(
|
53 |
+
negative_prompt=None,
|
54 |
+
prompt_embeds=pos_emb,
|
55 |
+
negative_prompt_embeds=neg_emb,
|
56 |
+
prompt_attention_mask=pos_ids.attention_mask,
|
57 |
+
negative_prompt_attention_mask=neg_ids.attention_mask,
|
58 |
+
max_sequence_length=512,
|
59 |
+
width=width,
|
60 |
+
height=height,
|
61 |
+
num_inference_steps=num_inference_steps,
|
62 |
+
generator=generator,
|
63 |
+
guidance_scale=4.5).images[0]
|
64 |
+
return image, seed
|
65 |
+
|
66 |
+
examples = [
|
67 |
+
"芝生の上にあるピザ",
|
68 |
+
"東京の桜と建物。満開の桜の木が並び、ピンク色の花びらが風に舞っている。桜の背景には東京の高層ビルや伝統的な建物が調和して立っている。春の陽光が全体を明るく照らし、桜と建物が美しく映えている。都市の活気と自然の美しさが融合した風景。",
|
69 |
+
"パリは燃えているか",
|
70 |
+
]
|
71 |
+
|
72 |
+
css="""
|
73 |
+
#col-container {
|
74 |
+
margin: 0 auto;
|
75 |
+
max-width: 520px;
|
76 |
+
}
|
77 |
+
"""
|
78 |
+
|
79 |
+
with gr.Blocks(css=css) as demo:
|
80 |
+
|
81 |
+
with gr.Column(elem_id="col-container"):
|
82 |
+
gr.Markdown(f"""# CommonArt β
|
83 |
+
商用利用できる透明性の高い日本語画像生成AI
|
84 |
+
""")
|
85 |
+
|
86 |
+
with gr.Row():
|
87 |
+
|
88 |
+
prompt = gr.Text(
|
89 |
+
label="テキスト",
|
90 |
+
show_label=False,
|
91 |
+
max_lines=1,
|
92 |
+
placeholder="生成したいものを日本語や英語で説明してください",
|
93 |
+
container=False,
|
94 |
+
)
|
95 |
+
|
96 |
+
run_button = gr.Button("生成", scale=0)
|
97 |
+
|
98 |
+
result = gr.Image(label="生成結果", show_label=False)
|
99 |
+
|
100 |
+
with gr.Accordion("詳細設定", open=False):
|
101 |
+
|
102 |
+
seed = gr.Slider(
|
103 |
+
label="シード値",
|
104 |
+
minimum=0,
|
105 |
+
maximum=MAX_SEED,
|
106 |
+
step=1,
|
107 |
+
value=0,
|
108 |
+
)
|
109 |
+
|
110 |
+
randomize_seed = gr.Checkbox(label="ランダム", value=True)
|
111 |
+
|
112 |
+
with gr.Row():
|
113 |
+
|
114 |
+
width = gr.Slider(
|
115 |
+
label="幅",
|
116 |
+
minimum=256,
|
117 |
+
maximum=768,
|
118 |
+
step=64,
|
119 |
+
value=512,
|
120 |
+
)
|
121 |
+
|
122 |
+
height = gr.Slider(
|
123 |
+
label="高さ",
|
124 |
+
minimum=256,
|
125 |
+
maximum=768,
|
126 |
+
step=64,
|
127 |
+
value=512,
|
128 |
+
)
|
129 |
+
|
130 |
+
with gr.Row():
|
131 |
+
|
132 |
+
|
133 |
+
num_inference_steps = gr.Slider(
|
134 |
+
label="推論回数",
|
135 |
+
minimum=1,
|
136 |
+
maximum=50,
|
137 |
+
step=1,
|
138 |
+
value=20,
|
139 |
+
)
|
140 |
+
|
141 |
+
gr.Examples(
|
142 |
+
examples = examples,
|
143 |
+
fn = infer,
|
144 |
+
inputs = [prompt],
|
145 |
+
outputs = [result, seed],
|
146 |
+
cache_examples="lazy"
|
147 |
+
)
|
148 |
+
|
149 |
+
gr.on(
|
150 |
+
triggers=[run_button.click, prompt.submit],
|
151 |
+
fn = infer,
|
152 |
+
inputs = [prompt, seed, randomize_seed, width, height, num_inference_steps],
|
153 |
+
outputs = [result, seed]
|
154 |
+
)
|
155 |
+
|
156 |
+
demo.launch()
|