File size: 6,539 Bytes
6267052 524b414 a6ad525 524b414 a6ad525 6267052 a6ad525 f05f013 a6ad525 524b414 47ab942 f05f013 47ab942 a6ad525 f05f013 a6ad525 f05f013 6267052 f05f013 524b414 f05f013 a6ad525 f05f013 6267052 f05f013 524b414 f05f013 524b414 6267052 524b414 ffe4561 524b414 ffe4561 524b414 903a60a 26eb541 cdc565e 26eb541 f7fa496 cdc565e f7fa496 903a60a 26eb541 f7fa496 903a60a cdc565e |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 |
#import spaces
import random
import gradio as gr
import numpy as np
import PIL.Image
import torch
from compel import Compel, ReturnedEmbeddingsType
from diffusers import (EulerAncestralDiscreteScheduler,
StableDiffusionXLPipeline)
from PIL import Image
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
DEFAULT_BATCH_SIZE = 2
MAX_BATCH_SIZE = 12
# Make sure to use torch.float16 consistently throughout the pipeline
pipe = StableDiffusionXLPipeline.from_pretrained(
"dhead/wai-nsfw-illustrious-sdxl-v140-sdxl",
torch_dtype=torch.float16,
variant="fp16", # Explicitly use fp16 variant
use_safetensors=True # Use safetensors if available
)
pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(pipe.scheduler.config)
pipe.to(device)
# Force all components to use the same dtype
pipe.text_encoder.to(torch.float16)
pipe.text_encoder_2.to(torch.float16)
pipe.vae.to(torch.float16)
pipe.unet.to(torch.float16)
# 追加: Initialize Compel for long prompt processing
compel = Compel(
tokenizer=[pipe.tokenizer, pipe.tokenizer_2],
text_encoder=[pipe.text_encoder, pipe.text_encoder_2],
returned_embeddings_type=ReturnedEmbeddingsType.PENULTIMATE_HIDDEN_STATES_NON_NORMALIZED,
requires_pooled=[False, True],
truncate_long_prompts=False
)
MAX_SEED = np.iinfo(np.int32).max
MAX_IMAGE_SIZE = 2000
# 追加: Simple long prompt processing function
def process_long_prompt(prompt, negative_prompt=""):
try:
conditioning, pooled = compel([prompt, negative_prompt])
return conditioning, pooled
except Exception as e:
print(f"Long prompt processing failed: {e}")
return None, None
def generate_image(
prompt,
negative_prompt,
generator,
width,
height,
guidance_scale,
num_inference_steps,
):
return pipe(
prompt=prompt,
negative_prompt=negative_prompt,
guidance_scale=guidance_scale,
num_inference_steps=num_inference_steps,
width=width,
height=height,
generator=generator,
).images[0]
#@spaces.GPU
def infer(
prompt,
negative_prompt,
seed,
randomize_seed,
batch_size,
width,
height,
guidance_scale,
num_inference_steps,
):
if randomize_seed:
seed = random.randint(0, MAX_SEED)
prompts = [prompt] * batch_size
negative_prompts = [negative_prompt] * batch_size
generators = [
torch.Generator(device=device).manual_seed(seed - i)
for i in range(batch_size)
]
try:
images = pipe(
prompt=prompts,
negative_prompt=negative_prompts,
guidance_scale=guidance_scale,
num_inference_steps=num_inference_steps,
width=width,
height=height,
generator=generators,
).images
return images
except RuntimeError as e:
print(f"Generation error: {e}")
return [Image.new("RGB", (width, height), (0, 0, 0))] * batch_size
css = """
#col-container {
margin: 0 auto;
max-width: 2000px;
}
"""
with gr.Blocks(css=css) as demo:
with gr.Accordion("Advanced Settings", open=False):
negative_prompt = gr.Text(
label="Negative prompt",
max_lines=5,
placeholder="Enter a negative prompt",
value=" ((unimaginative, dry, flat, static, stiff, uninspired), bad quality overexposed, too bright, washed out, high exposure, low resolution, artifact, compression artifacts, low poly, blocky, banding, color bleed, texture seams, oversaturation, fused fingers, malformed eyes, missing iris sclera, poorly drawn background)"
)
seed = gr.Slider(
label="Seed",
minimum=0,
maximum=MAX_SEED,
step=20,
value=0,
)
randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
with gr.Row():
width = gr.Slider(
label="Width",
minimum=256,
maximum=MAX_IMAGE_SIZE,
step=32,
value=600,
)
height = gr.Slider(
label="Height",
minimum=256,
maximum=MAX_IMAGE_SIZE,
step=32,
value=800,
)
with gr.Row():
guidance_scale = gr.Slider(
label="Guidance scale",
minimum=0.0,
maximum=30.0,
step=0.1,
value=7,
)
batch_size = gr.Slider(
label="Batch size",
minimum=1,
maximum=MAX_BATCH_SIZE,
step=1,
value=DEFAULT_BATCH_SIZE,
)
with gr.Row():
num_inference_steps = gr.Slider(
label="Number of inference steps",
minimum=1,
maximum=99,
step=1,
value=20,
)
with gr.Column(elem_id="col-container"):
with gr.Row():
with gr.Row():
prompt = gr.Text(
label="Prompt",
show_label=False,
max_lines=16,
placeholder="Enter your prompt (long prompts are automatically supported)",
value="Show me a glorious mountain range covered in colorful crystaline trees with a laser disco show from space.",
container=False,
)
results = [
gr.Image(format="png", show_label=False)
for _ in range(MAX_BATCH_SIZE)
]
run_button = gr.Button("Run", scale=0)
run_button.click(
concurrency_limit=25,
fn=infer,
inputs=[
prompt,
negative_prompt,
seed,
randomize_seed,
batch_size,
width,
height,
guidance_scale,
num_inference_steps,
],
outputs=results,
)
with gr.Column():
for _ in range(5):
with gr.Row():
for _ in range(2):
results.append(gr.Image(format="png", show_label=False, visible=True))
demo.queue(default_concurrency_limit=25).launch()
|