Spaces:
Running
on
Zero
Running
on
Zero
File size: 7,105 Bytes
ad6ae44 e85a61d 5fa0aa7 e85a61d 7c7890b cac33a7 ad6ae44 7c7890b cac33a7 f6d7fe3 d52c01d cac33a7 595db11 5fa0aa7 cac33a7 0290645 5fa0aa7 7c7890b e85a61d 7c7890b 0290645 5fa0aa7 7c7890b 259d09b 595db11 cac33a7 0290645 259d09b ad6ae44 cac33a7 5fa0aa7 259d09b cac33a7 259d09b cac33a7 e85a61d 0290645 cac33a7 7c7890b 0290645 e85a61d 0290645 7c7890b ad6ae44 7c7890b ad6ae44 7c7890b ad6ae44 7c7890b 259d09b ad6ae44 ebd6aa7 7c7890b 0290645 f6d7fe3 d52c01d 7c7890b ad6ae44 7c7890b ad6ae44 7c7890b 4e4bf12 259d09b 4e4bf12 7c7890b ad6ae44 818225c ad6ae44 f6d7fe3 ad6ae44 f6d7fe3 7c7890b ad6ae44 7c7890b |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 |
import spaces
from diffusers import (
StableDiffusionXLPipeline,
EulerDiscreteScheduler,
UNet2DConditionModel,
AutoencoderTiny,
)
import torch
import os
from huggingface_hub import hf_hub_download
from compel import Compel, ReturnedEmbeddingsType
from gradio_promptweighting import PromptWeighting
from PIL import Image
import gradio as gr
import time
from safetensors.torch import load_file
import time
import tempfile
from pathlib import Path
# Constants
BASE = "stabilityai/stable-diffusion-xl-base-1.0"
REPO = "ByteDance/SDXL-Lightning"
# 1-step
CHECKPOINT = "sdxl_lightning_2step_unet.safetensors"
taesd_model = "madebyollin/taesdxl"
SFAST_COMPILE = os.environ.get("SFAST_COMPILE", "0") == "1"
SAFETY_CHECKER = os.environ.get("SAFETY_CHECKER", "0") == "1"
USE_TAESD = os.environ.get("USE_TAESD", "0") == "1"
# check if MPS is available OSX only M1/M2/M3 chips
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
torch_device = device
torch_dtype = torch.float16
print(f"SAFETY_CHECKER: {SAFETY_CHECKER}")
print(f"SFAST_COMPILE: {SFAST_COMPILE}")
print(f"USE_TAESD: {USE_TAESD}")
print(f"device: {device}")
unet = UNet2DConditionModel.from_config(BASE, subfolder="unet").to(
"cuda", torch.float16
)
unet.load_state_dict(load_file(hf_hub_download(REPO, CHECKPOINT), device="cuda"))
pipe = StableDiffusionXLPipeline.from_pretrained(
BASE, unet=unet, torch_dtype=torch.float16, variant="fp16", safety_checker=False
).to("cuda")
unet = unet.to(dtype=torch.float16)
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],
)
if USE_TAESD:
pipe.vae = AutoencoderTiny.from_pretrained(
taesd_model, torch_dtype=torch_dtype, use_safetensors=True
).to(device)
# Ensure sampler uses "trailing" timesteps.
pipe.scheduler = EulerDiscreteScheduler.from_config(
pipe.scheduler.config, timestep_spacing="trailing"
)
pipe.set_progress_bar_config(disable=True)
if SAFETY_CHECKER:
from safety_checker import StableDiffusionSafetyChecker
from transformers import CLIPFeatureExtractor
safety_checker = StableDiffusionSafetyChecker.from_pretrained(
"CompVis/stable-diffusion-safety-checker"
).to(device)
feature_extractor = CLIPFeatureExtractor.from_pretrained(
"openai/clip-vit-base-patch32"
)
def check_nsfw_images(
images: list[Image.Image],
) -> tuple[list[Image.Image], list[bool]]:
safety_checker_input = feature_extractor(images, return_tensors="pt").to(device)
has_nsfw_concepts = safety_checker(
images=[images],
clip_input=safety_checker_input.pixel_values.to(torch_device),
)
return images, has_nsfw_concepts
if SFAST_COMPILE:
from sfast.compilers.diffusion_pipeline_compiler import compile, CompilationConfig
# sfast compilation
config = CompilationConfig.Default()
try:
import xformers
config.enable_xformers = True
except ImportError:
print("xformers not installed, skip")
try:
import triton
config.enable_triton = True
except ImportError:
print("Triton not installed, skip")
# CUDA Graph is suggested for small batch sizes and small resolutions to reduce CPU overhead.
# But it can increase the amount of GPU memory used.
# For StableVideoDiffusionPipeline it is not needed.
config.enable_cuda_graph = True
pipe = compile(pipe, config)
@spaces.GPU
def predict(prompt, prompt_w, guidance_scale, seed=1231231):
generator = torch.manual_seed(seed)
last_time = time.time()
prompt_w = " ".join(
[f"({p['prompt']}){p['scale']}" for p in prompt_w if p["prompt"]]
)
conditioning, pooled = compel([prompt + " " + prompt_w, ""])
results = pipe(
prompt_embeds=conditioning[0:1],
pooled_prompt_embeds=pooled[0:1],
negative_prompt_embeds=conditioning[1:2],
negative_pooled_prompt_embeds=pooled[1:2],
generator=generator,
num_inference_steps=2,
guidance_scale=guidance_scale,
# width=768,
# height=768,
output_type="pil",
)
print(f"Pipe took {time.time() - last_time} seconds")
if SAFETY_CHECKER:
images, has_nsfw_concepts = check_nsfw_images(results.images)
if any(has_nsfw_concepts):
gr.Warning("NSFW content detected.")
return Image.new("RGB", (512, 512))
image = results.images[0]
with tempfile.NamedTemporaryFile(suffix=".jpg", delete=False) as tmpfile:
image.save(tmpfile, "JPEG", quality=80, optimize=True, progressive=True)
return Path(tmpfile.name)
css = """
#container{
margin: 0 auto;
max-width: 80rem;
}
#intro{
max-width: 100%;
margin: 0 auto;
}
.generating {
display: none
}
"""
with gr.Blocks(css=css) as demo:
with gr.Column(elem_id="container"):
gr.Markdown(
"""
# SDXL-Lightning- Text To Image 2-Steps
**Model**: https://huggingface.co/ByteDance/SDXL-Lightning
""",
elem_id="intro",
)
with gr.Row():
with gr.Column():
with gr.Group():
prompt = gr.Textbox(
placeholder="Insert your prompt here:",
max_lines=1,
label="Prompt",
)
prompt_w = PromptWeighting(
min=0,
max=3,
step=0.005,
show_label=False,
info="Drag up and down to adjust the weight of each prompt.",
)
with gr.Accordion("Advanced options", open=True):
seed = gr.Slider(
minimum=0,
maximum=12013012031030,
label="Seed",
step=1,
)
guidance_scale = gr.Slider(
minimum=0.0,
maximum=20.0,
label="Guidance scale",
value=0.0,
step=0.1,
)
generate_bt = gr.Button("Generate")
with gr.Column():
image = gr.Image(type="filepath")
inputs = [
prompt,
prompt_w,
guidance_scale,
seed,
]
outputs = [image]
gr.on(
triggers=[
prompt.input,
prompt_w.input,
generate_bt.click,
guidance_scale.input,
seed.input,
],
fn=predict,
inputs=inputs,
outputs=outputs,
show_progress="hidden",
show_api=False,
trigger_mode="always_last",
)
demo.queue(api_open=False)
demo.launch()
|