File size: 4,923 Bytes
4120479
ba844d5
c694422
4120479
1475e41
ba844d5
5042a41
ba844d5
4120479
ba844d5
 
4120479
ad8076c
3eb8dac
5042a41
ba844d5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8428948
ba844d5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ce04d24
ba844d5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
de0fd08
ba844d5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f494526
ba844d5
 
 
 
 
 
 
f494526
ba844d5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b4852dc
ba844d5
 
b6fa736
ba844d5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import gradio as gr
from huggingface_hub import login
from diffusers import DiffusionPipeline, StableDiffusionXLPipeline
import torch
import copy
import os
import spaces
import random

hf_token = os.environ.get("HF_TOKEN")
login(token = hf_token)

original_pipe = DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16)

@spaces.GPU
def infer(lora_1_id, lora_1_sfts, lora_2_id, lora_2_sfts, prompt, negative_prompt, lora_1_scale, lora_2_scale, seed):

    unet = copy.deepcopy(original_pipe.unet)
    text_encoder = copy.deepcopy(original_pipe.text_encoder)
    text_encoder_2 = copy.deepcopy(original_pipe.text_encoder_2)

    pipe = StableDiffusionXLPipeline(
        vae = original_pipe.vae,
        text_encoder = text_encoder,
        text_encoder_2 = text_encoder_2,
        scheduler = original_pipe.scheduler,
        tokenizer = original_pipe.tokenizer,
        tokenizer_2 = original_pipe.tokenizer_2,
        unet = unet
    )

    pipe.to("cuda")

    pipe.load_lora_weights(
        lora_1_id,
        weight_name = lora_1_sfts,        
        low_cpu_mem_usage = True,
        use_auth_token = True
    )

    pipe.fuse_lora(lora_1_scale)

    pipe.load_lora_weights(
        lora_2_id,
        weight_name = lora_2_sfts,        
        low_cpu_mem_usage = True,
        use_auth_token = True
    )

    pipe.fuse_lora(lora_2_scale)

    if negative_prompt == "" :
        negative_prompt = None
    
    if seed < 0 :
        seed = random.randit(0, 423538377342)
    
    generator = torch.Generator(device="cuda").manual_seed(seed)

    image = pipe(
        prompt = prompt,
        negative_prompt = negative_prompt,
        num_inference_steps = 25,
        width = 1024,
        height = 1024,
        generator = generator
    ).images[0]
    
    return image, seed

with gr.Blocks() as demo:
    with gr.Column(elem_id="col-container"):

        title = gr.HTML(
        '''
        <h1 style="text-align: center;">LoRA Fusion</h1>
        <p style="text-align: center;">Fuse 2 custom LoRa models</p>
        '''
        )
        
        # PART 1 • MODELS
        with gr.Row():
            
            with gr.Column():
                
                lora_1_id = gr.Textbox(
                    label = "LoRa 1 ID",
                    placeholder = "username/model_id"
                )

                lora_1_sfts = gr.Textbox(
                    label = "Safetensors file",
                    placeholder = "specific_chosen.safetensors"
                )
            
            with gr.Column():
                
                lora_2_id = gr.Textbox(
                    label = "LoRa 2 ID",
                    placeholder = "username/model_id"
                )

                lora_2_sfts = gr.Textbox(
                    label = "Safetensors file",
                    placeholder = "specific_chosen.safetensors"
                )

        # PART 2 • INFERENCE
        with gr.Row():
            
            prompt = gr.Textbox(
                label = "Your prompt",
                info = "Use your trigger words into a coherent prompt",
                placeholde = "e.g: a triggerWordOne portrait in triggerWord2 style"
            )

            run_btn = gr.Button("Run")
        
        output_image = gr.Image(
            label = "Output"
        )

        # Advanced Settings
        with gr.Accordion("Advanced Settings", open=False):
            
            with gr.Row():
                
                lora_1_scale = gr.Slider(
                    label = "LoRa 1 scale",
                    minimum = 0,
                    maximum = 1,
                    step = 0.1,
                    value = 0.7
                )
                
                lora_2_scale = gr.Slider(
                    label = "LoRa 2 scale",
                    minimum = 0,
                    maximum = 1,
                    step = 0.1,
                    value = 0.7
                )
            
            negative_prompt = gr.Textbox(
                label = "Negative prompt"
            )

            seed = gr.Slider(
                label = "Seed",
                info = "-1 denotes a random seed",
                minimum = -1,
                maximum = 423538377342,
                value = -1
            )

            last_used_seed = gr.Number(
                label = "Last used seed",
                info = "the seed used in the last generation",
            )
    
    # ACTIONS
    run_btn.click(
        fn = infer,
        inputs = [
            lora_1_id,
            lora_1_sfts,
            lora_2_id,
            lora_2_sfts,
            prompt,
            negative_prompt,
            lora_1_scale,
            lora_2_scale,
            seed
        ],
        outputs = [
            output_image, 
            last_used_seed
        ]
    )

demo.queue(concurrency_count=2).launch()