Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -1,6 +1,6 @@
|
|
1 |
from flask import Flask, request, jsonify
|
2 |
from flask_cors import CORS
|
3 |
-
from diffusers import StableDiffusionPipeline,
|
4 |
import torch
|
5 |
import os
|
6 |
from PIL import Image
|
@@ -9,106 +9,114 @@ import time
|
|
9 |
from accelerate import Accelerator
|
10 |
import logging
|
11 |
|
12 |
-
|
|
|
|
|
|
|
|
|
13 |
CORS(app)
|
14 |
|
15 |
-
|
16 |
logging.basicConfig(level=logging.DEBUG, format='%(asctime)s - %(levelname)s - %(message)s')
|
17 |
logger = logging.getLogger(__name__)
|
18 |
|
19 |
-
|
20 |
-
accelerator = Accelerator(
|
21 |
|
22 |
-
|
23 |
model_cache = {}
|
24 |
model_paths = {
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
|
29 |
-
|
30 |
ratio_to_dims = {
|
31 |
-
|
32 |
-
|
33 |
-
|
34 |
-
|
35 |
|
36 |
def load_model(model_id):
|
37 |
-
|
38 |
-
|
39 |
-
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
|
44 |
-
|
45 |
-
|
46 |
-
|
47 |
-
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
-
|
56 |
-
|
|
|
|
|
57 |
|
58 |
@app.route('/')
|
59 |
def index():
|
60 |
-
|
61 |
|
62 |
@app.route('/generate', methods=['POST'])
|
63 |
def generate():
|
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 |
if __name__ == '__main__':
|
114 |
-
|
|
|
1 |
from flask import Flask, request, jsonify
|
2 |
from flask_cors import CORS
|
3 |
+
from diffusers import StableDiffusionPipeline, DPMSolverMultistepScheduler
|
4 |
import torch
|
5 |
import os
|
6 |
from PIL import Image
|
|
|
9 |
from accelerate import Accelerator
|
10 |
import logging
|
11 |
|
12 |
+
# Disable GPU detection
|
13 |
+
os.environ["CUDA_VISIBLE_DEVICES"] = ""
|
14 |
+
torch.set_default_device("cpu")
|
15 |
+
|
16 |
+
app = Flask(__name__, static_folder='static')
|
17 |
CORS(app)
|
18 |
|
19 |
+
# Configure logging
|
20 |
logging.basicConfig(level=logging.DEBUG, format='%(asctime)s - %(levelname)s - %(message)s')
|
21 |
logger = logging.getLogger(__name__)
|
22 |
|
23 |
+
# Initialize Accelerator for CPU
|
24 |
+
accelerator = Accelerator(device_placement=False)
|
25 |
|
26 |
+
# Model cache
|
27 |
model_cache = {}
|
28 |
model_paths = {
|
29 |
+
"ssd-1b": "remiai3/ssd-1b",
|
30 |
+
"sd-v1-5": "remiai3/stable-diffusion-v1-5"
|
31 |
+
}
|
32 |
|
33 |
+
# Image ratio to dimensions (optimized for CPU)
|
34 |
ratio_to_dims = {
|
35 |
+
"1:1": (256, 256),
|
36 |
+
"3:4": (192, 256),
|
37 |
+
"16:9": (256, 144)
|
38 |
+
}
|
39 |
|
40 |
def load_model(model_id):
|
41 |
+
if model_id not in model_cache:
|
42 |
+
logger.info(f"Loading model {model_id}...")
|
43 |
+
try:
|
44 |
+
pipe = StableDiffusionPipeline.from_pretrained(
|
45 |
+
model_paths[model_id],
|
46 |
+
torch_dtype=torch.float32,
|
47 |
+
use_auth_token=os.getenv("HF_TOKEN"),
|
48 |
+
use_safetensors=True,
|
49 |
+
low_cpu_mem_usage=True,
|
50 |
+
device_map="cpu"
|
51 |
+
)
|
52 |
+
pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config)
|
53 |
+
pipe = accelerator.prepare(pipe)
|
54 |
+
pipe.enable_attention_slicing()
|
55 |
+
pipe.enable_sequential_cpu_offload()
|
56 |
+
pipe.to("cpu")
|
57 |
+
model_cache[model_id] = pipe
|
58 |
+
logger.info(f"Model {model_id} loaded successfully")
|
59 |
+
except Exception as e:
|
60 |
+
logger.error(f"Error loading model {model_id}: {str(e)}")
|
61 |
+
raise
|
62 |
+
return model_cache[model_id]
|
63 |
|
64 |
@app.route('/')
|
65 |
def index():
|
66 |
+
return app.send_static_file('index.html')
|
67 |
|
68 |
@app.route('/generate', methods=['POST'])
|
69 |
def generate():
|
70 |
+
try:
|
71 |
+
data = request.json
|
72 |
+
model_id = data.get('model', 'ssd-1b')
|
73 |
+
prompt = data.get('prompt', '')
|
74 |
+
ratio = data.get('ratio', '1:1')
|
75 |
+
num_images = min(int(data.get('num_images', 1)), 4)
|
76 |
+
guidance_scale = float(data.get('guidance_scale', 7.5))
|
77 |
+
|
78 |
+
if not prompt:
|
79 |
+
return jsonify({"error": "Prompt is required"}), 400
|
80 |
+
|
81 |
+
if model_id == 'ssd-1b' and num_images > 1:
|
82 |
+
return jsonify({"error": "SSD-1B allows only 1 image per generation"}), 400
|
83 |
+
if model_id == 'ssd-1b' and ratio != '1:1':
|
84 |
+
return jsonify({"error": "SSD-1B supports only 1:1 ratio"}), 400
|
85 |
+
if model_id == 'sd-v1-5' and len(prompt.split()) > 77:
|
86 |
+
return jsonify({"error": "Prompt exceeds 77 tokens for Stable Diffusion v1.5"}), 400
|
87 |
+
|
88 |
+
width, height = ratio_to_dims.get(ratio, (256, 256))
|
89 |
+
pipe = load_model(model_id)
|
90 |
+
pipe.to("cpu")
|
91 |
+
|
92 |
+
images = []
|
93 |
+
num_inference_steps = 20 if model_id == 'ssd-1b' else 30
|
94 |
+
for _ in range(num_images):
|
95 |
+
image = pipe(
|
96 |
+
prompt=prompt,
|
97 |
+
height=height,
|
98 |
+
width=width,
|
99 |
+
num_inference_steps=num_inference_steps,
|
100 |
+
guidance_scale=guidance_scale
|
101 |
+
).images[0]
|
102 |
+
images.append(image)
|
103 |
+
|
104 |
+
output_dir = "outputs"
|
105 |
+
os.makedirs(output_dir, exist_ok=True)
|
106 |
+
image_urls = []
|
107 |
+
for i, img in enumerate(images):
|
108 |
+
img_path = os.path.join(output_dir, f"generated_{int(time.time())}_{i}.png")
|
109 |
+
img.save(img_path)
|
110 |
+
with open(img_path, "rb") as f:
|
111 |
+
img_data = base64.b64encode(f.read()).decode('utf-8')
|
112 |
+
image_urls.append(f"data:image/png;base64,{img_data}")
|
113 |
+
os.remove(img_path)
|
114 |
+
|
115 |
+
return jsonify({"images": image_urls})
|
116 |
+
|
117 |
+
except Exception as e:
|
118 |
+
logger.error(f"Image generation failed: {str(e)}")
|
119 |
+
return jsonify({"error": f"Image generation failed: {str(e)}"}), 500
|
120 |
|
121 |
if __name__ == '__main__':
|
122 |
+
app.run(host='0.0.0.0', port=7860)
|