A fine-tuned Stable Diffusion Inpainting model for virtual jewelry try-on. This model enables users to virtually try on necklaces and other jewelry items on person images.

Model Description

Naari Jewelry VTON is designed for the jewelry virtual try-on task. It uses inpainting techniques to seamlessly place jewelry items onto images of people, creating realistic visualizations of how jewelry would look when worn.

Key Features

Virtual try-on for necklaces and jewelry
Based on Stable Diffusion Inpainting architecture
Trained on jewelry segmentation datasets
Supports mask-guided inpainting

Intended Uses

E-commerce: Allow customers to virtually try jewelry before purchasing
Fashion Apps: Enable jewelry visualization in mobile applications
Content Creation: Generate realistic jewelry wearing images

How to Use

python

from diffusers import StableDiffusionInpaintPipeline
import torch
from PIL import Image

# Load the model
pipe = StableDiffusionInpaintPipeline.from_pretrained(
    "GaneshGowri/naari-jewelry-vton",
    torch_dtype=torch.float16
)
pipe = pipe.to("cuda")

# Load your images
image = Image.open("person_image.png").convert("RGB")
mask = Image.open("neck_mask.png").convert("RGB")

# Generate
result = pipe(
    prompt="elegant gold necklace on person",
    image=image,
    mask_image=mask,
    num_inference_steps=50,
    guidance_scale=7.5
).images[0]

result.save("result.png")

Training Details

Base Model: runwayml/stable-diffusion-inpainting
Training Data: VITON-HD dataset + Roboflow Necklace Segmentation
Training Platform: Kaggle (T4x2 GPU)
Framework: Hugging Face Diffusers

Limitations

Best results with front-facing portrait images
Mask quality affects output quality
May require prompt tuning for specific jewelry types

Citation

bibtex

@misc{naari-jewelry-vton,
  author = {GaneshGowri},
  title = {Naari Jewelry VTON},
  year = {2026},
  publisher = {Hugging Face},
  url = {https://huggingface.co/GaneshGowri/naari-jewelry-vton}
}

License

This model is released under the Apache 2.0 license.