PhotoDoodle: Turn Photos into Art with Just Text Prompts 106

What is PhotoDoodle?

PhotoDoodle ai is an image editing system jointly developed by ByteDance and research teams from Chinese and Singaporean universities, built upon the Flux.1 model. The highlight of this technology is its ability to learn and imitate specific artistic styles from a small number of samples, and then transform users' photos into works in various artistic styles.

Features of PhotoDoodle

• Intelligent Style Transfer: Using an EditLoRA variant, it can learn an individual artist's style from just a few examples. By studying selected image pairs and delving deeply into the nuances of each artistic style, this technology can apply different artistic styles to users' photos.

• Position Encoding Cloning Technology: The AI can remember the exact position of every pixel in the original image. When adding new elements, it can use the stored positional information to place them precisely and seamlessly blend them into the image, without the need for additional parameter training, greatly improving processing efficiency and enabling highly customized image processing.

• Efficient Learning Capability: Built on the Flux.1 model, it can quickly learn new artistic styles or editing techniques, reducing reliance on large amounts of data.

• Creative Expression tool: Provides a platform for both artists and non-professional users to create artistic-looking works without needing complex image editing skills.

Applications of PhotoDoodle

• Personal Creative Sharing: Users can transform everyday photos into artistic works and share them on social media.

• Advertising and Design: Brands and designers can use PhotoDoodle to quickly generate visual content with specific styles for advertising campaigns or product design.

• Education and Teaching: In art and design courses, it serves as a teaching tool to help students understand the transformation process between different artistic styles.

• Personalized Product Creation: For example, converting personal photos into artistic-style decorative paintings, phone cases, and more.

How to Use PhotoDoodle

Dataset and Open-Source Code:

• The research team has released a dataset containing six different artistic styles and over 300 image pairs, along with code provided on GitHub. Users can visit the PhotoDoodle GitHub repository for more information.

Operating Steps:

• Users can use the provided code and dataset to train the model for personalized style editing, or directly use the pre-trained model for image editing.

PhotoDoodle GitHub: https://github.com/showlab/PhotoDoodle

PhotoDoodle HuggingFace: https://huggingface.co/datasets/nicolaus-huang/PhotoDoodle