File size: 6,522 Bytes
c70f6b6
 
 
 
8c6f026
0cea845
c70f6b6
 
 
 
 
 
d071a59
16eb805
 
 
d071a59
c70f6b6
 
bc71736
 
 
 
 
 
d071a59
9102b04
 
 
 
bc71736
 
d071a59
 
bc71736
d071a59
16eb805
 
 
 
 
 
d071a59
9102b04
 
 
bc71736
9102b04
 
d071a59
9102b04
 
bc71736
d071a59
9102b04
 
 
bc71736
9102b04
 
bc71736
 
 
 
1d15b14
9102b04
 
c70f6b6
16eb805
bc71736
 
 
 
 
 
 
 
 
d071a59
 
 
16eb805
 
c70f6b6
16eb805
bc71736
 
 
 
 
 
 
 
 
 
16eb805
bc71736
16eb805
 
bc71736
16eb805
c70f6b6
 
 
 
 
 
 
 
 
 
 
9102b04
c70f6b6
 
 
 
9102b04
 
 
 
 
 
 
 
 
 
 
ce1c2ea
c70f6b6
 
 
 
 
 
 
 
 
 
 
d071a59
c70f6b6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
bc71736
c70f6b6
 
 
 
 
 
 
bc71736
c70f6b6
 
 
 
 
 
 
8e62d24
 
 
c70f6b6
 
 
 
 
 
 
 
 
 
9102b04
 
 
 
 
 
 
 
 
 
 
 
 
c70f6b6
 
9102b04
 
 
c70f6b6
 
 
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
import gradio as gr
import numpy as np
import spaces
import torch
import random
from peft import PeftModel
from diffusers import FluxControlPipeline, FluxTransformer2DModel
from image_gen_aux import DepthPreprocessor

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

# Initialize models without moving to CUDA yet
pipe = FluxControlPipeline.from_pretrained(
    "black-forest-labs/FLUX.1-Depth-dev", 
    torch_dtype=torch.bfloat16
)
processor = DepthPreprocessor.from_pretrained("LiheYoung/depth-anything-large-hf")

def cleanup_memory():
    """Clean up GPU memory"""
    if torch.cuda.is_available():
        torch.cuda.empty_cache()
        torch.cuda.ipc_collect()

@spaces.GPU
def load_lora(lora_path):
    if not lora_path.strip():
        return "Please provide a valid LoRA path"
    try:
        cleanup_memory()
        
        # Move to GPU within the wrapped function
        pipe.to("cuda")
        pipe.enable_model_cpu_offload()
        
        # Unload any existing LoRA weights first
        try:
            pipe.unload_lora_weights()
        except:
            pass
            
        # Load new LoRA weights
        pipe.load_lora_weights(lora_path)
        return f"Successfully loaded LoRA weights from {lora_path}"
    except Exception as e:
        cleanup_memory()
        return f"Error loading LoRA weights: {str(e)}"

@spaces.GPU
def unload_lora():
    try:
        cleanup_memory()
        pipe.to("cuda")
        pipe.unload_lora_weights()
        return "Successfully unloaded LoRA weights"
    except Exception as e:
        cleanup_memory()
        return f"Error unloading LoRA weights: {str(e)}"

def round_to_multiple(number, multiple):
    """Round a number to the nearest multiple"""
    return multiple * round(number / multiple)

@spaces.GPU
def infer(control_image, prompt, seed=42, randomize_seed=False, width=1024, height=1024, 
          guidance_scale=3.5, num_inference_steps=28, progress=gr.Progress(track_tqdm=True)):
    
    try:
        cleanup_memory()
        
        if randomize_seed:
            seed = random.randint(0, MAX_SEED)
        
        # Ensure dimensions are divisible by 16
        width = round_to_multiple(width, 16)
        height = round_to_multiple(height, 16)
        
        # Move pipeline to GPU within the wrapped function
        pipe.to("cuda")
        
        # Process control image
        control_image = processor(control_image)[0].convert("RGB")
        
        # Generate image
        with torch.inference_mode():
            image = pipe(
                prompt=prompt,
                control_image=control_image,
                height=height,
                width=width,
                num_inference_steps=num_inference_steps,
                guidance_scale=guidance_scale,
                generator=torch.Generator("cuda").manual_seed(seed),
            ).images[0]
        
        cleanup_memory()
        return image, seed
    except Exception as e:
        cleanup_memory()
        return None, f"Error during inference: {str(e)}"

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 Depth [dev] with LoRA Support
12B param rectified flow transformer structural conditioning tuned, guidance-distilled from [FLUX.1 [pro]](https://blackforestlabs.ai/)  
[[non-commercial license](https://huggingface.co/black-forest-labs/FLUX.1-dev/blob/main/LICENSE.md)] [[blog](https://blackforestlabs.ai/announcing-black-forest-labs/)] [[model](https://huggingface.co/black-forest-labs/FLUX.1-dev)]
        """)

        # LoRA controls
        with gr.Row():
            lora_path = gr.Textbox(
                label="HuggingFace LoRA Path",
                placeholder="e.g., Borcherding/FLUX.1-dev-LoRA-AutumnSpringTrees"
            )
            load_lora_btn = gr.Button("Load LoRA")
            unload_lora_btn = gr.Button("Unload LoRA")
        
        lora_status = gr.Textbox(label="LoRA Status", interactive=False)

        control_image = gr.Image(label="Upload the image for control", type="pil")
        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)
        error_message = gr.Textbox(label="Error", visible=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=16,  # Changed to 16 to ensure divisibility
                    value=1024,
                )
                
                height = gr.Slider(
                    label="Height",
                    minimum=256,
                    maximum=MAX_IMAGE_SIZE,
                    step=16,  # Changed to 16 to ensure divisibility
                    value=1024,
                )
            
            with gr.Row():
                guidance_scale = gr.Slider(
                    label="Guidance Scale",
                    minimum=1,
                    maximum=30,
                    step=0.5,
                    value=10,
                )
  
                num_inference_steps = gr.Slider(
                    label="Number of inference steps",
                    minimum=1,
                    maximum=50,
                    step=1,
                    value=28,
                )

    # Event handlers
    load_lora_btn.click(
        fn=load_lora,
        inputs=[lora_path],
        outputs=[lora_status]
    )
    
    unload_lora_btn.click(
        fn=unload_lora,
        inputs=[],
        outputs=[lora_status]
    )

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

demo.launch()