File size: 6,862 Bytes
b8975f7
 
 
 
 
e984a2c
 
b8975f7
b9a20a2
b8975f7
87a083e
e984a2c
085c378
 
f5e7ff1
 
 
085c378
b8975f7
085c378
f5e7ff1
085c378
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b8975f7
085c378
f5e7ff1
 
 
 
 
 
 
 
 
 
 
e04fd6c
b8975f7
 
 
 
37a00e0
3cb91bd
 
b8975f7
37a00e0
b8975f7
085c378
 
b8975f7
 
b6a9837
085c378
f5e7ff1
085c378
f5e7ff1
085c378
 
 
 
 
 
 
 
f5e7ff1
085c378
 
 
 
 
 
 
 
 
 
 
 
 
b8975f7
 
 
 
3ce81f3
b8975f7
c996b7d
7492d5d
c53d852
085c378
c53d852
3ce81f3
 
039e573
085c378
039e573
cb7c885
 
 
039e573
085c378
b8975f7
085c378
b8975f7
3ce81f3
 
 
 
 
 
 
 
 
664d54d
b8975f7
085c378
 
 
 
 
 
 
f5e7ff1
 
 
 
 
 
 
 
 
 
b8975f7
 
 
 
 
 
 
 
 
 
 
7ab092f
f5e7ff1
 
3cb91bd
b8975f7
 
 
7ab092f
f5e7ff1
 
3cb91bd
3ce81f3
b8975f7
 
 
 
f5e7ff1
b8975f7
f5e7ff1
 
 
 
 
 
 
 
3ce81f3
 
b8975f7
 
 
7f10b7b
b8975f7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f5e7ff1
b8975f7
085c378
b8975f7
 
f5e7ff1
3ce81f3
 
b8975f7
e04fd6c
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
import os
import random
import uuid
import json

import gradio as gr
import numpy as np
from PIL import Image
import spaces
import torch
from diffusers import StableDiffusionXLPipeline, EulerAncestralDiscreteScheduler

# Use environment variables for flexibility
MODEL_ID = os.getenv("MODEL_ID", "sd-community/sdxl-flash")
MAX_IMAGE_SIZE = int(os.getenv("MAX_IMAGE_SIZE", "4096"))
USE_TORCH_COMPILE = os.getenv("USE_TORCH_COMPILE", "0") == "1"
ENABLE_CPU_OFFLOAD = os.getenv("ENABLE_CPU_OFFLOAD", "0") == "1"
BATCH_SIZE = int(os.getenv("BATCH_SIZE", "1"))  # Allow generating multiple images at once

# Determine device and load model outside of function for efficiency
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
pipe = StableDiffusionXLPipeline.from_pretrained(
    MODEL_ID,
    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
    use_safetensors=True,
    add_watermarker=False,
).to(device)
pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(pipe.scheduler.config)

# Torch compile for potential speedup (experimental)
if USE_TORCH_COMPILE:
    pipe.compile()

# CPU offloading for larger RAM capacity (experimental)
if ENABLE_CPU_OFFLOAD:
    pipe.enable_model_cpu_offload()

MAX_SEED = np.iinfo(np.int32).max

def save_image(img):
    unique_name = str(uuid.uuid4()) + ".png"
    img.save(unique_name)
    return unique_name

def randomize_seed_fn(seed: int, randomize_seed: bool) -> int:
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)
    return seed

@spaces.GPU(duration=35, enable_queue=True)
def generate(
    prompt: str,
    negative_prompt: str = "",
    use_negative_prompt: bool = False,
    seed: int = 1,
    width: int = 1024,
    height: int = 1024,
    guidance_scale: float = 3,
    num_inference_steps: int = 30,
    randomize_seed: bool = False,
    use_resolution_binning: bool = True, 
    num_images: int = 1,  # Number of images to generate
    progress=gr.Progress(track_tqdm=True),
):
    seed = int(randomize_seed_fn(seed, randomize_seed))
    generator = torch.Generator(device=device).manual_seed(seed)

    # Improved options handling
    options = {
        "prompt": [prompt] * num_images,
        "negative_prompt": [negative_prompt] * num_images if use_negative_prompt else None,
        "width": width,
        "height": height,
        "guidance_scale": guidance_scale,
        "num_inference_steps": num_inference_steps,
        "generator": generator,
        "output_type": "pil",
    }

    # Use resolution binning for faster generation with less VRAM usage
    if use_resolution_binning:
        options["use_resolution_binning"] = True

    # Generate images potentially in batches
    images = []
    for i in range(0, num_images, BATCH_SIZE):
        batch_options = options.copy()
        batch_options["prompt"] = options["prompt"][i:i+BATCH_SIZE]
        if "negative_prompt" in batch_options:
            batch_options["negative_prompt"] = options["negative_prompt"][i:i+BATCH_SIZE]
        images.extend(pipe(**batch_options).images)

    image_paths = [save_image(img) for img in images]
    return image_paths, seed

examples = [
    "a cat eating a piece of cheese",
    "a ROBOT riding a BLUE horse on Mars, photorealistic, 4k",
    "Ironman VS Hulk, ultrarealistic",
    "Astronaut in a jungle, cold color palette, oil pastel, detailed, 8k",
    "An alien holding a sign board containing the word 'Flash', futuristic, neonpunk",
    "Kids going to school, Anime style"
]

css = '''
.gradio-container{max-width: 700px !important}
h1{text-align:center}
footer {
    visibility: hidden
}
'''

with gr.Blocks(css=css) as demo:
    gr.Markdown("""# SDXL Flash""")
    with gr.Group():
        with gr.Row():
            prompt = gr.Text(
                label="Prompt",
                show_label=False,
                max_lines=1,
                placeholder="Enter your prompt",
                container=False,
            )
            run_button = gr.Button("Run", scale=0)
        result = gr.Gallery(label="Result", columns=1, show_label=False) 
    with gr.Accordion("Advanced options", open=False):
        num_images = gr.Slider(
            label="Number of Images",
            minimum=1,
            maximum=4,
            step=1,
            value=1,
        )
        with gr.Row():
            use_negative_prompt = gr.Checkbox(label="Use negative prompt", value=True)
            negative_prompt = gr.Text(
                label="Negative prompt",
                max_lines=5,
                lines=4,
                placeholder="Enter a negative prompt",
                value="(deformed, distorted, disfigured:1.3), poorly drawn, bad anatomy, wrong anatomy, extra limb, missing limb, floating limbs, (mutated hands and fingers:1.4), disconnected limbs, mutation, mutated, ugly, disgusting, blurry, amputation, NSFW",
                visible=True,
            )
        seed = gr.Slider(
            label="Seed",
            minimum=0,
            maximum=MAX_SEED,
            step=1,
            value=0,
        )
        randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
        with gr.Row(visible=True):
            width = gr.Slider(
                label="Width",
                minimum=512,
                maximum=MAX_IMAGE_SIZE,
                step=64,
                value=1024,
            )
            height = gr.Slider(
                label="Height",
                minimum=512,
                maximum=MAX_IMAGE_SIZE,
                step=64,
                value=1024,
            )
        with gr.Row():
            guidance_scale = gr.Slider(
                label="Guidance Scale",
                minimum=0.1,
                maximum=6,
                step=0.1,
                value=3.0,
            )
            num_inference_steps = gr.Slider(
                label="Number of inference steps",
                minimum=1,
                maximum=15,
                step=1,
                value=8,
            )

    gr.Examples(
        examples=examples,
        inputs=prompt,
        cache_examples=False
    )

    use_negative_prompt.change(
        fn=lambda x: gr.update(visible=x),
        inputs=use_negative_prompt,
        outputs=negative_prompt,
        api_name=False,
    )

    gr.on(
        triggers=[
            prompt.submit,
            negative_prompt.submit,
            run_button.click,
        ],
        fn=generate,
        inputs=[
            prompt,
            negative_prompt,
            use_negative_prompt,
            seed,
            width,
            height,
            guidance_scale,
            num_inference_steps,
            randomize_seed,
            num_images
        ],
        outputs=[result, seed],
        api_name="run",
    )

if __name__ == "__main__":
    demo.queue(max_size=50).launch()