File size: 4,876 Bytes
7f891bb
4e6d911
 
 
2e306db
a7d057d
d9f1205
4e6d911
 
8a31b39
 
 
6f5f495
8a31b39
 
 
 
 
 
 
6f5f495
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7f891bb
4e6d911
 
9d86930
d2cb214
4e6d911
 
 
8a31b39
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1c4aefd
8a31b39
 
4e6d911
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1c4aefd
4e6d911
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d53ee34
 
d2b0012
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
import gradio as gr
import numpy as np
import random
import spaces
import torch
from diffusers import DiffusionPipeline

device = "cuda" if torch.cuda.is_available() else "cpu"

# Load the model in FP16
pipe = DiffusionPipeline.from_pretrained("black-forest-labs/FLUX.1-schnell", torch_dtype=torch.float16)

# Move the pipeline to GPU if available
pipe = pipe.to(device)

# Convert text encoders to full precision
pipe.text_encoder = pipe.text_encoder.to(torch.float32)
if hasattr(pipe, 'text_encoder_2'):
    pipe.text_encoder_2 = pipe.text_encoder_2.to(torch.float32)

# Enable memory efficient attention if available and on CUDA
if device == "cuda" and hasattr(pipe, 'enable_xformers_memory_efficient_attention'):
    try:
        pipe.enable_xformers_memory_efficient_attention()
        print("xformers memory efficient attention enabled")
    except Exception as e:
        print(f"Could not enable memory efficient attention: {e}")

# Compile the UNet for potential speedups if on CUDA
if device == "cuda":
    try:
        pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)
        print("UNet compiled for potential speedups")
    except Exception as e:
        print(f"Could not compile UNet: {e}")

MAX_SEED = np.iinfo(np.int32).max
MAX_IMAGE_SIZE = 2048

@spaces.GPU()
def infer(prompt, seed=42, randomize_seed=False, width=1024, height=1024, num_inference_steps=4, progress=gr.Progress(track_tqdm=True)):
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)
    generator = torch.Generator(device=device).manual_seed(seed)
    
    # Use full precision for text encoding
    with torch.no_grad():
        text_inputs = pipe.tokenizer(prompt, return_tensors="pt").to(device)
        text_embeddings = pipe.text_encoder(text_inputs.input_ids)[0]
    
    # Use mixed precision for the rest of the pipeline
    with torch.inference_mode(), torch.autocast(device_type='cuda', dtype=torch.float16):
        image = pipe(
            prompt_embeds=text_embeddings,
            width=width,
            height=height,
            num_inference_steps=num_inference_steps,
            generator=generator,
            guidance_scale=0.0
        ).images[0]
    
    return image, seed
 
examples = [
    "a tiny astronaut hatching from an egg on the moon",
    "a cat holding a sign that says hello world",
    "an anime illustration of a wiener schnitzel",
]

css="""
#col-container {
    margin: 0 auto;
    max-width: 520px;
}
"""

with gr.Blocks(css=css) as demo:
    
    with gr.Column(elem_id="col-container"):
        gr.Markdown(f"""# FLUX.1 [schnell]
12B param rectified flow transformer distilled from [FLUX.1 [pro]](https://blackforestlabs.ai/) for 4 step generation
[[blog](https://blackforestlabs.ai/announcing-black-forest-labs/)] [[model](https://huggingface.co/black-forest-labs/FLUX.1-schnell)]
        """)
        
        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.Image(label="Result", show_label=False)
        
        with gr.Accordion("Advanced Settings", open=False):
            
            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():
                
                width = gr.Slider(
                    label="Width",
                    minimum=256,
                    maximum=MAX_IMAGE_SIZE,
                    step=32,
                    value=1024,
                )
                
                height = gr.Slider(
                    label="Height",
                    minimum=256,
                    maximum=MAX_IMAGE_SIZE,
                    step=32,
                    value=1024,
                )
            
            with gr.Row():
                
  
                num_inference_steps = gr.Slider(
                    label="Number of inference steps",
                    minimum=1,
                    maximum=50,
                    step=1,
                    value=4,
                )
        
        gr.Examples(
            examples = examples,
            fn = infer,
            inputs = [prompt],
            outputs = [result, seed],
            cache_examples="lazy"
        )

    gr.on(
        triggers=[run_button.click, prompt.submit],
        fn = infer,
        inputs = [prompt, seed, randomize_seed, width, height, num_inference_steps],
        outputs = [result, seed]
    )

demo.launch()