Spaces:
Running
Running
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,79 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
from diffusers import StableDiffusionControlNetPipeline, ControlNetModel, UniPCMultistepScheduler
|
3 |
+
from PIL import Image, ImageFilter
|
4 |
+
import numpy as np
|
5 |
+
import gradio as gr
|
6 |
+
import cv2
|
7 |
+
|
8 |
+
# Load pre-trained Stable Diffusion model (frozen part)
|
9 |
+
model_id = "runwayml/stable-diffusion-v1-5"
|
10 |
+
controlnet_id = "lllyasviel/control_v11p_sd15_canny" # ControlNet for edge detection-based control
|
11 |
+
|
12 |
+
# Load ControlNet model (trainable part)
|
13 |
+
controlnet = ControlNetModel.from_pretrained(controlnet_id, torch_dtype=torch.float16)
|
14 |
+
|
15 |
+
# Load Stable Diffusion pipeline with ControlNet
|
16 |
+
pipe = StableDiffusionControlNetPipeline.from_pretrained(
|
17 |
+
model_id, controlnet=controlnet, torch_dtype=torch.float16
|
18 |
+
)
|
19 |
+
|
20 |
+
# Use an efficient scheduler
|
21 |
+
pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config)
|
22 |
+
|
23 |
+
# Move pipeline to GPU
|
24 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
25 |
+
pipe.to(device)
|
26 |
+
|
27 |
+
# Function to generate control image (edge detection using Canny filter)
|
28 |
+
def generate_control_image(input_image_path):
|
29 |
+
image = cv2.imread(input_image_path, cv2.IMREAD_GRAYSCALE)
|
30 |
+
edges = cv2.Canny(image, 100, 200) # Apply Canny edge detection
|
31 |
+
control_image = Image.fromarray(edges).convert("L")
|
32 |
+
control_image = control_image.resize((512, 512)) # Resize to match model requirements
|
33 |
+
control_image.save("control_image.jpg")
|
34 |
+
return "control_image.jpg"
|
35 |
+
|
36 |
+
# Function to apply color change
|
37 |
+
def apply_color_change(input_image, prompt):
|
38 |
+
# Save input image temporarily
|
39 |
+
input_image_path = "input_image.jpg"
|
40 |
+
input_image.save(input_image_path)
|
41 |
+
|
42 |
+
# Generate control image (edges)
|
43 |
+
control_image_path = generate_control_image(input_image_path)
|
44 |
+
|
45 |
+
# Load processed input and control images
|
46 |
+
input_image = Image.open(input_image_path).convert("RGB").resize((512, 512))
|
47 |
+
control_image = Image.open(control_image_path).convert("L")
|
48 |
+
|
49 |
+
# Generate the new image using the pipeline
|
50 |
+
generator = torch.manual_seed(42) # For reproducibility
|
51 |
+
output_image = pipe(
|
52 |
+
prompt=prompt,
|
53 |
+
image=input_image,
|
54 |
+
control_image=control_image,
|
55 |
+
generator=generator,
|
56 |
+
num_inference_steps=30
|
57 |
+
).images[0]
|
58 |
+
|
59 |
+
output_image.save("output_color_changed.png")
|
60 |
+
return "output_color_changed.png"
|
61 |
+
|
62 |
+
# Gradio interface
|
63 |
+
def gradio_interface(input_image, prompt):
|
64 |
+
output_image_path = apply_color_change(input_image, prompt)
|
65 |
+
return output_image_path
|
66 |
+
|
67 |
+
# Launch the Gradio interface with drag and drop
|
68 |
+
interface = gr.Interface(
|
69 |
+
fn=gradio_interface,
|
70 |
+
inputs=[
|
71 |
+
gr.Image(type="pil", label="Upload your image"), # Drag and drop feature
|
72 |
+
gr.Textbox(label="Enter prompt", placeholder="e.g. A hoodie with blue and white design"),
|
73 |
+
],
|
74 |
+
outputs=gr.Image(label="Color Changed Output"),
|
75 |
+
title="AI-Powered Clothing Color Changer",
|
76 |
+
description="Upload an image of clothing, enter a prompt, and get a redesigned color version.",
|
77 |
+
)
|
78 |
+
|
79 |
+
interface.launch()
|