init
Browse files- app.py +57 -60
- requirements.txt +1 -5
app.py
CHANGED
@@ -1,26 +1,25 @@
|
|
1 |
-
|
2 |
-
|
3 |
-
import spaces # Necessary for the @spaces.GPU decorator
|
4 |
-
from diffusers import DiffusionPipeline, EulerAncestralDiscreteScheduler, EulerDiscreteScheduler
|
5 |
-
import torch
|
6 |
import os
|
|
|
7 |
from datetime import datetime
|
8 |
from PIL import Image
|
9 |
import boto3
|
10 |
from botocore.exceptions import NoCredentialsError
|
11 |
from dotenv import load_dotenv
|
|
|
12 |
|
13 |
# Carregar variáveis de ambiente do arquivo .env
|
14 |
load_dotenv()
|
15 |
|
16 |
-
# AWS S3
|
17 |
AWS_ACCESS_KEY = os.getenv('AWS_ACCESS_KEY')
|
18 |
AWS_SECRET_KEY = os.getenv('AWS_SECRET_KEY')
|
19 |
AWS_BUCKET_NAME = os.getenv('AWS_BUCKET_NAME')
|
20 |
AWS_REGION = os.getenv('AWS_REGION')
|
21 |
-
HF_TOKEN = os.getenv('HF_TOKEN') #
|
22 |
|
23 |
-
#
|
24 |
s3_client = boto3.client(
|
25 |
's3',
|
26 |
aws_access_key_id=AWS_ACCESS_KEY,
|
@@ -28,102 +27,100 @@ s3_client = boto3.client(
|
|
28 |
region_name=AWS_REGION
|
29 |
)
|
30 |
|
31 |
-
#
|
32 |
character_pipe = DiffusionPipeline.from_pretrained(
|
33 |
"cagliostrolab/animagine-xl-3.1",
|
34 |
torch_dtype=torch.float16,
|
35 |
use_safetensors=True,
|
36 |
-
use_auth_token=HF_TOKEN #
|
37 |
)
|
38 |
character_pipe.scheduler = EulerDiscreteScheduler.from_config(character_pipe.scheduler.config)
|
39 |
|
40 |
-
#
|
41 |
item_pipe = DiffusionPipeline.from_pretrained(
|
42 |
"openart-custom/DynaVisionXL",
|
43 |
torch_dtype=torch.float16,
|
44 |
use_safetensors=True,
|
45 |
-
use_auth_token=HF_TOKEN #
|
46 |
)
|
47 |
item_pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(item_pipe.scheduler.config)
|
48 |
|
49 |
-
#
|
50 |
-
@spaces.GPU(duration=60) #
|
51 |
def generate_image(model_type, prompt, negative_prompt, width, height, guidance_scale, num_inference_steps):
|
52 |
if model_type == "character":
|
53 |
pipe = character_pipe
|
54 |
default_prompt = "1girl, souji okita, fate series, solo, upper body, bedroom, night, seducing, (sexy clothes)"
|
55 |
-
default_negative_prompt = "lowres, (bad), text, error, fewer, extra, missing, worst quality, jpeg artifacts,
|
|
|
|
|
56 |
elif model_type == "item":
|
57 |
pipe = item_pipe
|
58 |
default_prompt = "great sword, runes on blade, acid on blade, weapon, (((item)))"
|
59 |
default_negative_prompt = "1girl, girl, man, boy, 1man, men, girls"
|
60 |
else:
|
61 |
-
return "
|
62 |
|
63 |
-
#
|
64 |
final_prompt = prompt if prompt else default_prompt
|
65 |
final_negative_prompt = negative_prompt if negative_prompt else default_negative_prompt
|
66 |
|
67 |
-
# Move
|
68 |
pipe.to("cuda")
|
69 |
|
70 |
-
#
|
71 |
-
|
72 |
prompt=final_prompt,
|
73 |
negative_prompt=final_negative_prompt,
|
74 |
width=int(width),
|
75 |
height=int(height),
|
76 |
guidance_scale=float(guidance_scale),
|
77 |
num_inference_steps=int(num_inference_steps)
|
78 |
-
)
|
|
|
79 |
|
80 |
-
#
|
81 |
temp_file = "/tmp/generated_image.png"
|
82 |
image.save(temp_file)
|
83 |
|
84 |
-
#
|
85 |
file_name = datetime.now().strftime("%Y%m%d_%H%M%S") + ".png"
|
86 |
try:
|
87 |
s3_client.upload_file(temp_file, AWS_BUCKET_NAME, file_name)
|
88 |
s3_url = f"https://{AWS_BUCKET_NAME}.s3.{AWS_REGION}.amazonaws.com/{file_name}"
|
89 |
return s3_url
|
90 |
except NoCredentialsError:
|
91 |
-
return "
|
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 |
-
if result is None:
|
123 |
-
raise HTTPException(status_code=400, detail="Invalid input")
|
124 |
-
return {"result": result}
|
125 |
|
126 |
-
# Run the FastAPI app with Uvicorn
|
127 |
if __name__ == "__main__":
|
128 |
-
|
129 |
-
uvicorn.run(app, host="0.0.0.0", port=7860)
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import spaces # Necessário para o decorador @spaces.GPU (caso esteja usando Hugging Face Spaces)
|
|
|
|
|
|
|
3 |
import os
|
4 |
+
import torch
|
5 |
from datetime import datetime
|
6 |
from PIL import Image
|
7 |
import boto3
|
8 |
from botocore.exceptions import NoCredentialsError
|
9 |
from dotenv import load_dotenv
|
10 |
+
from diffusers import DiffusionPipeline, EulerAncestralDiscreteScheduler, EulerDiscreteScheduler
|
11 |
|
12 |
# Carregar variáveis de ambiente do arquivo .env
|
13 |
load_dotenv()
|
14 |
|
15 |
+
# Configurações do AWS S3
|
16 |
AWS_ACCESS_KEY = os.getenv('AWS_ACCESS_KEY')
|
17 |
AWS_SECRET_KEY = os.getenv('AWS_SECRET_KEY')
|
18 |
AWS_BUCKET_NAME = os.getenv('AWS_BUCKET_NAME')
|
19 |
AWS_REGION = os.getenv('AWS_REGION')
|
20 |
+
HF_TOKEN = os.getenv('HF_TOKEN') # Token da Hugging Face
|
21 |
|
22 |
+
# Inicializar cliente S3
|
23 |
s3_client = boto3.client(
|
24 |
's3',
|
25 |
aws_access_key_id=AWS_ACCESS_KEY,
|
|
|
27 |
region_name=AWS_REGION
|
28 |
)
|
29 |
|
30 |
+
# Configuração do pipeline para "character"
|
31 |
character_pipe = DiffusionPipeline.from_pretrained(
|
32 |
"cagliostrolab/animagine-xl-3.1",
|
33 |
torch_dtype=torch.float16,
|
34 |
use_safetensors=True,
|
35 |
+
use_auth_token=HF_TOKEN # Inclui o token aqui
|
36 |
)
|
37 |
character_pipe.scheduler = EulerDiscreteScheduler.from_config(character_pipe.scheduler.config)
|
38 |
|
39 |
+
# Configuração do pipeline para "item"
|
40 |
item_pipe = DiffusionPipeline.from_pretrained(
|
41 |
"openart-custom/DynaVisionXL",
|
42 |
torch_dtype=torch.float16,
|
43 |
use_safetensors=True,
|
44 |
+
use_auth_token=HF_TOKEN # Inclui o token aqui
|
45 |
)
|
46 |
item_pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(item_pipe.scheduler.config)
|
47 |
|
48 |
+
# Função de geração de imagem com alocação de GPU (através do decorador do Hugging Face Spaces)
|
49 |
+
@spaces.GPU(duration=60) # Aloca a GPU somente durante a execução desta função
|
50 |
def generate_image(model_type, prompt, negative_prompt, width, height, guidance_scale, num_inference_steps):
|
51 |
if model_type == "character":
|
52 |
pipe = character_pipe
|
53 |
default_prompt = "1girl, souji okita, fate series, solo, upper body, bedroom, night, seducing, (sexy clothes)"
|
54 |
+
default_negative_prompt = ("lowres, (bad), text, error, fewer, extra, missing, worst quality, jpeg artifacts, "
|
55 |
+
"low quality, watermark, unfinished, displeasing, oldest, early, chromatic aberration, "
|
56 |
+
"signature, extra digits, artistic error, username, scan, [abstract]")
|
57 |
elif model_type == "item":
|
58 |
pipe = item_pipe
|
59 |
default_prompt = "great sword, runes on blade, acid on blade, weapon, (((item)))"
|
60 |
default_negative_prompt = "1girl, girl, man, boy, 1man, men, girls"
|
61 |
else:
|
62 |
+
return "Tipo inválido. Escolha entre 'character' ou 'item'."
|
63 |
|
64 |
+
# Se o usuário fornecer prompt, utiliza-o; caso contrário, usa o padrão
|
65 |
final_prompt = prompt if prompt else default_prompt
|
66 |
final_negative_prompt = negative_prompt if negative_prompt else default_negative_prompt
|
67 |
|
68 |
+
# Move o pipeline para a GPU
|
69 |
pipe.to("cuda")
|
70 |
|
71 |
+
# Geração da imagem
|
72 |
+
result = pipe(
|
73 |
prompt=final_prompt,
|
74 |
negative_prompt=final_negative_prompt,
|
75 |
width=int(width),
|
76 |
height=int(height),
|
77 |
guidance_scale=float(guidance_scale),
|
78 |
num_inference_steps=int(num_inference_steps)
|
79 |
+
)
|
80 |
+
image = result.images[0]
|
81 |
|
82 |
+
# Salva a imagem em um arquivo temporário
|
83 |
temp_file = "/tmp/generated_image.png"
|
84 |
image.save(temp_file)
|
85 |
|
86 |
+
# Faz upload para o AWS S3
|
87 |
file_name = datetime.now().strftime("%Y%m%d_%H%M%S") + ".png"
|
88 |
try:
|
89 |
s3_client.upload_file(temp_file, AWS_BUCKET_NAME, file_name)
|
90 |
s3_url = f"https://{AWS_BUCKET_NAME}.s3.{AWS_REGION}.amazonaws.com/{file_name}"
|
91 |
return s3_url
|
92 |
except NoCredentialsError:
|
93 |
+
return "Credenciais não disponíveis"
|
94 |
+
|
95 |
+
# Função que integra a geração via Gradio
|
96 |
+
def gradio_generate(model_type, prompt, negative_prompt, width, height, guidance_scale, num_inference_steps):
|
97 |
+
return generate_image(model_type, prompt, negative_prompt, width, height, guidance_scale, num_inference_steps)
|
98 |
+
|
99 |
+
# Definindo os componentes de entrada utilizando a API atual do Gradio
|
100 |
+
model_type_input = gr.Dropdown(choices=["character", "item"], value="character", label="Model Type")
|
101 |
+
prompt_input = gr.Textbox(lines=2, placeholder="Digite o prompt (deixe vazio para o padrão)", label="Prompt")
|
102 |
+
negative_prompt_input = gr.Textbox(lines=2, placeholder="Digite o negative prompt (deixe vazio para o padrão)", label="Negative Prompt")
|
103 |
+
width_input = gr.Number(value=512, label="Width")
|
104 |
+
height_input = gr.Number(value=512, label="Height")
|
105 |
+
guidance_scale_input = gr.Number(value=7.5, label="Guidance Scale")
|
106 |
+
num_inference_steps_input = gr.Number(value=50, label="Number of Inference Steps")
|
107 |
+
|
108 |
+
# Criação da interface Gradio
|
109 |
+
iface = gr.Interface(
|
110 |
+
fn=gradio_generate,
|
111 |
+
inputs=[
|
112 |
+
model_type_input,
|
113 |
+
prompt_input,
|
114 |
+
negative_prompt_input,
|
115 |
+
width_input,
|
116 |
+
height_input,
|
117 |
+
guidance_scale_input,
|
118 |
+
num_inference_steps_input,
|
119 |
+
],
|
120 |
+
outputs="text",
|
121 |
+
title="Image Generation API",
|
122 |
+
description="Gere imagens usando modelos de difusão e faça upload para o AWS S3."
|
123 |
+
)
|
|
|
|
|
|
|
124 |
|
|
|
125 |
if __name__ == "__main__":
|
126 |
+
iface.launch()
|
|
requirements.txt
CHANGED
@@ -1,10 +1,6 @@
|
|
1 |
-
|
2 |
-
uvicorn
|
3 |
-
transformers
|
4 |
-
spaces
|
5 |
diffusers
|
6 |
torch
|
7 |
boto3
|
8 |
python-dotenv
|
9 |
Pillow
|
10 |
-
accelerate
|
|
|
1 |
+
gradio
|
|
|
|
|
|
|
2 |
diffusers
|
3 |
torch
|
4 |
boto3
|
5 |
python-dotenv
|
6 |
Pillow
|
|