I have used the fashn vton code to change saree among different models. The code is given as:

Importing Required Packages

from fashn_vton import TryOnPipeline
from PIL import Image
import numpy as np

The Function to get the VITON result

def get_viton_result(model_image: str, cloth_image: str):

garment = Image.open(cloth_image).convert("RGB")
target_w, target_h = garment.size
person = Image.open(model_image).convert("RGB").resize((target_w, target_h), Image.Resampling.LANCZOS)

# Initialize pipeline (auto-detects GPU)
pipeline = TryOnPipeline(weights_dir = r"C:\Users\Webbies\Jupyter_Notebooks\TITAN\Fashion_VITON\fashn-vton-1.5\weights")

# Run inference
result = pipeline(
    person_image = person,
    garment_image = garment,
    category = "one-pieces",
    garment_photo_type = "model",
    num_timesteps = 50,
    guidance_scale = 1.5,
    seed = 42
)

# Save output
image = result.images[0].resize((target_w, target_h), Image.Resampling.LANCZOS)
image_np = np.array(image)

return image_np

The slim_bg is the slim person (model in black saree). the image_1 (the blue saree with black border) is the saree image and the remaining image is the result. I want to know why the result image is generated as this?

There are gaps in the result image from which the background is shown. The gaps does not exsists in any of the person or cloth image. Any way to solve this issue?