STABLE-DIFF-CARTOON

Cartoonish Image Generation with Fine-Tuned Stable Diffusion Model

This repository contains the fine-tuned Stable Diffusion model for generating cartoonish images. The model has been fine-tuned using DreamBooth on top of SG161222/Realistic_Vision_V2.0 and is optimized to produce high-quality cartoon-style images.

Model Overview

• Model Name: Fine-Tuned Stable Diffusion for Cartoonish Images
• Base Model: SG161222/Realistic_Vision_V2.0
• Fine-Tuning Method: DreamBooth
• Output File: my_dreambooth_model.safetensors
• Input Image Size: 512x512 pixels

Why 512x512 Images?

The model has been trained and fine-tuned to work specifically with images of size 512x512 pixels. This resolution was chosen because it strikes a balance between image quality and computational efficiency, making it suitable for generating detailed images without requiring excessive GPU resources. Using images of this size ensures that the model can focus on generating high-quality features in the cartoon style.

Important: Preprocessing Your Dataset

To achieve the best results, ensure that all input images are resized to 512x512 pixels before fine-tuning or generating images. This step is crucial because the model expects inputs of this size, and mismatched dimensions can lead to poor quality or distorted outputs.

How to Use the Resizing Images Script:

1. Set the path to your original images and the directory where you want to save the 512x512 images.
2. Run the script to generate the 512x512 dataset.

Example usage:

!python scripts/512x512.py --input_dir /path/to/your/images --output_dir /path/to/save/512x512/images

Data Augmentation

Why Data Augmentation?

Data augmentation is an essential step in preparing your dataset for fine-tuning. It helps improve the model's robustness by exposing it to various transformations of the images, such as flipping, rotation, zooming, and color adjustments. This increased diversity helps the model generalize better to different styles and variations within the cartoonish theme.

Data Augmentation Script

A data augmentation script is provided in this repository under the scripts directory. The script applies various transformations to your dataset, ensuring that the model has a rich and diverse set of training images.

How to Use the Data Augmentation Script:

1. Set the path to your original images and the directory where you want to save the augmented images.
2. Adjust the augmentation operations if needed.
3. Run the script to generate the augmented dataset.

Example usage:

!python scripts/data_augmentation.py --input_dir /path/to/your/images --output_dir /path/to/save/augmented/images

Fine-Tuning the Model

Prerequisites

• A Google Colab environment with GPU enabled
• Installed dependencies: 'diffusers', 'torch', 'bitsandbytes'

Fine-Tuning & Testing Scripts

Use the provided Google Colab Notebook to fine-tune the model and then test the fine-tuned model . You can find the notebook here. The fine-tuned model will be saved as a '.safetensors' file in your Google Drive.

Dataset

The dataset used for fine-tuning is hosted on Kaggle. You can access and download it using the following link: Dataset