Spaces:
Running
on
Zero
Running
on
Zero
着手增加逻辑
Browse files
app.py
CHANGED
@@ -1,11 +1,20 @@
|
|
1 |
-
import
|
|
|
2 |
import gradio as gr
|
3 |
import numpy as np
|
4 |
import torch
|
|
|
|
|
5 |
import random
|
6 |
-
import
|
7 |
import utils
|
|
|
|
|
|
|
8 |
from diffusers.models import AutoencoderKL
|
|
|
|
|
|
|
9 |
from config import (
|
10 |
MODEL,
|
11 |
MIN_IMAGE_SIZE,
|
@@ -52,6 +61,37 @@ else:
|
|
52 |
pipe = None
|
53 |
|
54 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
55 |
|
56 |
@spaces.GPU
|
57 |
def generate(
|
@@ -70,23 +110,39 @@ def generate(
|
|
70 |
):
|
71 |
if randomize_seed:
|
72 |
seed = random.randint(0, MAX_SEED)
|
|
|
|
|
|
|
|
|
|
|
73 |
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
|
78 |
-
# negative_prompt=negative_prompt,
|
79 |
-
# guidance_scale=guidance_scale,
|
80 |
-
# num_inference_steps=num_inference_steps,
|
81 |
-
# width=width,
|
82 |
-
# height=height,
|
83 |
-
# generator=generator,
|
84 |
-
# ).images[0]
|
85 |
|
86 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
87 |
|
88 |
-
return None, seed
|
89 |
|
|
|
90 |
|
91 |
|
92 |
|
@@ -207,7 +263,7 @@ with gr.Blocks(css=custom_css).queue() as demo:
|
|
207 |
seed,randomize_seed,
|
208 |
guidance_scale,num_inference_steps
|
209 |
],
|
210 |
-
outputs=[result
|
211 |
)
|
212 |
|
213 |
if __name__ == "__main__":
|
|
|
1 |
+
import os
|
2 |
+
import gc
|
3 |
import gradio as gr
|
4 |
import numpy as np
|
5 |
import torch
|
6 |
+
import json
|
7 |
+
import spaces
|
8 |
import random
|
9 |
+
import config
|
10 |
import utils
|
11 |
+
import logging
|
12 |
+
from PIL import Image, PngImagePlugin
|
13 |
+
from datetime import datetime
|
14 |
from diffusers.models import AutoencoderKL
|
15 |
+
from diffusers import StableDiffusionXLPipeline, StableDiffusionXLImg2ImgPipeline
|
16 |
+
import time
|
17 |
+
from typing import List, Dict, Tuple, Optional
|
18 |
from config import (
|
19 |
MODEL,
|
20 |
MIN_IMAGE_SIZE,
|
|
|
61 |
pipe = None
|
62 |
|
63 |
|
64 |
+
class GenerationError(Exception):
|
65 |
+
"""Custom exception for generation errors"""
|
66 |
+
pass
|
67 |
+
|
68 |
+
def validate_prompt(prompt: str) -> str:
|
69 |
+
"""Validate and clean up the input prompt."""
|
70 |
+
if not isinstance(prompt, str):
|
71 |
+
raise GenerationError("Prompt must be a string")
|
72 |
+
try:
|
73 |
+
# Ensure proper UTF-8 encoding/decoding
|
74 |
+
prompt = prompt.encode('utf-8').decode('utf-8')
|
75 |
+
# Add space between ! and ,
|
76 |
+
prompt = prompt.replace("!,", "! ,")
|
77 |
+
except UnicodeError:
|
78 |
+
raise GenerationError("Invalid characters in prompt")
|
79 |
+
|
80 |
+
# Only check if the prompt is completely empty or only whitespace
|
81 |
+
if not prompt or prompt.isspace():
|
82 |
+
raise GenerationError("Prompt cannot be empty")
|
83 |
+
return prompt.strip()
|
84 |
+
|
85 |
+
def validate_dimensions(width: int, height: int) -> None:
|
86 |
+
"""Validate image dimensions."""
|
87 |
+
if not MIN_IMAGE_SIZE <= width <= MAX_IMAGE_SIZE:
|
88 |
+
raise GenerationError(f"Width must be between {MIN_IMAGE_SIZE} and {MAX_IMAGE_SIZE}")
|
89 |
+
|
90 |
+
if not MIN_IMAGE_SIZE <= height <= MAX_IMAGE_SIZE:
|
91 |
+
raise GenerationError(f"Height must be between {MIN_IMAGE_SIZE} and {MAX_IMAGE_SIZE}")
|
92 |
+
|
93 |
+
|
94 |
+
|
95 |
|
96 |
@spaces.GPU
|
97 |
def generate(
|
|
|
110 |
):
|
111 |
if randomize_seed:
|
112 |
seed = random.randint(0, MAX_SEED)
|
113 |
+
|
114 |
+
"""Generate images based on the given parameters."""
|
115 |
+
start_time = time.time()
|
116 |
+
upscaler_pipe = None
|
117 |
+
backup_scheduler = None
|
118 |
|
119 |
+
try:
|
120 |
+
# Memory management
|
121 |
+
torch.cuda.empty_cache()
|
122 |
+
gc.collect()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
123 |
|
124 |
+
return None
|
125 |
+
except GenerationError as e:
|
126 |
+
logger.warning(f"Generation validation error: {str(e)}")
|
127 |
+
raise gr.Error(str(e))
|
128 |
+
except Exception as e:
|
129 |
+
logger.exception("Unexpected error during generation")
|
130 |
+
raise gr.Error(f"Generation failed: {str(e)}")
|
131 |
+
finally:
|
132 |
+
# Cleanup
|
133 |
+
torch.cuda.empty_cache()
|
134 |
+
gc.collect()
|
135 |
+
|
136 |
+
if upscaler_pipe is not None:
|
137 |
+
del upscaler_pipe
|
138 |
+
|
139 |
+
if backup_scheduler is not None and pipe is not None:
|
140 |
+
pipe.scheduler = backup_scheduler
|
141 |
+
|
142 |
+
utils.free_memory()
|
143 |
|
|
|
144 |
|
145 |
+
|
146 |
|
147 |
|
148 |
|
|
|
263 |
seed,randomize_seed,
|
264 |
guidance_scale,num_inference_steps
|
265 |
],
|
266 |
+
outputs=[result],
|
267 |
)
|
268 |
|
269 |
if __name__ == "__main__":
|