File size: 5,190 Bytes
cac33a7
7c7890b
 
cac33a7
7c7890b
 
 
 
 
 
 
 
 
cac33a7
7c7890b
cac33a7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7c7890b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cac33a7
 
 
 
 
 
 
 
 
 
 
7c7890b
 
 
 
 
cac33a7
7c7890b
 
 
 
 
cac33a7
7c7890b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cac33a7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7c7890b
 
 
 
cac33a7
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
from diffusers import StableDiffusionXLPipeline, EulerDiscreteScheduler
import torch
import os
from huggingface_hub import hf_hub_download

try:
    import intel_extension_for_pytorch as ipex
except:
    pass

from PIL import Image
import gradio as gr
import time
from safetensors.torch import load_file


# Constants
BASE = "stabilityai/stable-diffusion-xl-base-1.0"
REPO = "ByteDance/SDXL-Lightning"
# 1-step
CHECKPOINT = "sdxl_lightning_1step_unet_x0.safetensors"

# {
#     "1-Step": ["sdxl_lightning_1step_unet_x0.safetensors", 1],
#     "2-Step": ["sdxl_lightning_2step_unet.safetensors", 2],
#     "4-Step": ["sdxl_lightning_4step_unet.safetensors", 4],
#     "8-Step": ["sdxl_lightning_8step_unet.safetensors", 8],
# }


TORCH_COMPILE = os.environ.get("TORCH_COMPILE", "0") == "1"
# check if MPS is available OSX only M1/M2/M3 chips
mps_available = hasattr(torch.backends, "mps") and torch.backends.mps.is_available()
xpu_available = hasattr(torch, "xpu") and torch.xpu.is_available()
device = torch.device(
    "cuda" if torch.cuda.is_available() else "xpu" if xpu_available else "cpu"
)
torch_device = device
torch_dtype = torch.float16

print(f"TORCH_COMPILE: {TORCH_COMPILE}")
print(f"device: {device}")

if mps_available:
    device = torch.device("mps")
    torch_device = "cpu"
    torch_dtype = torch.float32


pipe = StableDiffusionXLPipeline.from_pretrained(
    BASE, torch_dtype=torch.float16, variant="fp16"
)

pipe.scheduler = EulerDiscreteScheduler.from_config(
    pipe.scheduler.config, timestep_spacing="trailing", prediction_type="sample"
)

pipe.unet.load_state_dict(
    torch.load(load_file(hf_hub_download(REPO, CHECKPOINT)), map_location="cuda")
)

pipe.to(device=torch_device, dtype=torch_dtype).to(device)
pipe.set_progress_bar_config(disable=True)


def predict(prompt, seed=1231231):
    generator = torch.manual_seed(seed)
    last_time = time.time()
    results = pipe(
        prompt=prompt,
        generator=generator,
        num_inference_steps=1,
        guidance_scale=0.0,
        width=512,
        height=512,
        # original_inference_steps=params.lcm_steps,
        output_type="pil",
    )
    print(f"Pipe took {time.time() - last_time} seconds")
    nsfw_content_detected = (
        results.nsfw_content_detected[0]
        if "nsfw_content_detected" in results
        else False
    )
    if nsfw_content_detected:
        gr.Warning("NSFW content detected.")
        return Image.new("RGB", (512, 512))
    return results.images[0]


css = """
#container{
    margin: 0 auto;
    max-width: 40rem;
}
#intro{
    max-width: 100%;
    text-align: center;
    margin: 0 auto;
}
"""
with gr.Blocks(css=css) as demo:
    with gr.Column(elem_id="container"):
        gr.Markdown(
            """# SDXL Turbo - Text To Image
            ## Unofficial Demo
            SDXL Turbo model can generate high quality images in a single pass read more on [stability.ai post](https://stability.ai/news/stability-ai-sdxl-turbo).  
            **Model**: https://huggingface.co/stabilityai/sdxl-turbo
            """,
            elem_id="intro",
        )
        with gr.Row():
            with gr.Row():
                prompt = gr.Textbox(
                    placeholder="Insert your prompt here:", scale=5, container=False
                )
                generate_bt = gr.Button("Generate", scale=1)

        image = gr.Image(type="filepath")
        with gr.Accordion("Advanced options", open=False):
            seed = gr.Slider(
                randomize=True, minimum=0, maximum=12013012031030, label="Seed", step=1
            )
        with gr.Accordion("Run with diffusers"):
            gr.Markdown(
                """## Running SDXL Turbo with `diffusers`
            ```py
import torch
from diffusers import (
    StableDiffusionXLPipeline,
    UNet2DConditionModel,
    EulerDiscreteScheduler,
)
from huggingface_hub import hf_hub_download
from safetensors.torch import load_file

base = "stabilityai/stable-diffusion-xl-base-1.0"
repo = "ByteDance/SDXL-Lightning"
ckpt = "sdxl_lightning_1step_unet_x0.safetensors"  # Use the correct ckpt for your step setting!

# Load model.
unet = UNet2DConditionModel.from_config(base, subfolder="unet").to(
    "cuda", torch.float16
)
unet.load_state_dict(load_file(hf_hub_download(repo, ckpt), device="cuda"))
pipe = StableDiffusionXLPipeline.from_pretrained(
    base, unet=unet, torch_dtype=torch.float16, variant="fp16"
).to("cuda")

# Ensure sampler uses "trailing" timesteps and "sample" prediction type.
pipe.scheduler = EulerDiscreteScheduler.from_config(
    pipe.scheduler.config, timestep_spacing="trailing", prediction_type="sample"
)

# Ensure using the same inference steps as the loaded model and CFG set to 0.
pipe("A girl smiling", num_inference_steps=1, guidance_scale=0).images[0].save(
    "output.png"
)

            ```
            """
            )

        inputs = [prompt, seed]
        generate_bt.click(fn=predict, inputs=inputs, outputs=image, show_progress=False)
        prompt.input(fn=predict, inputs=inputs, outputs=image, show_progress=False)
        seed.change(fn=predict, inputs=inputs, outputs=image, show_progress=False)

demo.queue()
demo.launch()