PhotoMaker: Generate Person Photos Based on Descriptions 121

PhotoMaker uses multiple photos as identity IDs to capture people's features, then generates a new, personalized character image that matches a given description.

PhotoMaker can also blend the features of several different people's photos together to create an entirely new character. It can also change the gender and age of people in photos, and generate other photos in a variety of styles. It's fast, realistic, and produces natural-looking results.

PhotoMaker Main Features:

• Create photos from text descriptions: You can tell it what kind of person you want, and it will generate photos of people matching that description.

• Identity blending: If a user provides different image IDs as input, our PhotoMaker can integrate the features of these different IDs well to form a new ID.

• Change age or gender: For example, you can transform a photo of a man into a woman, or make the person in the photo look younger or older.

• Stylization: It can also add different artistic effects to photos, such as oil painting style or cartoon style, while preserving the ID attributes.

• Bring people from artwork/old photos into reality: Taking an artistic painting, sculpture, or old photo of a person as input, our PhotoMaker can "take a photo" of someone from the last century or even ancient times and bring them into the present century. We display the conditioning prompts below each image.

PhotoMaker Main Characteristics:

• Combining features from multiple photos: PhotoMaker can take multiple photos, which can be of the same person or different people. It analyzes the features in these photos and combines them to create a new, personalized character image.

• Flexible text control: Compared with other methods, PhotoMaker provides more flexible text control capabilities. Users can guide image generation through text prompts, which makes the produced images not only realistic but also better aligned with specific user requirements.

• High efficiency and fast generation: Compared with methods that require extensive fine-tuning (such as DreamBooth), PhotoMaker offers faster generation speed while maintaining identity information. It can generate customized human photos in a short time (for example, 10 seconds), which is very beneficial for application scenarios that require quick results.

• High identity fidelity: PhotoMaker better preserves a person's identity features when generating images. This is especially important for scenarios that require precise preservation of personal characteristics (such as personalized portraits or identity recognition).

How Does PhotoMaker Work?

PhotoMaker first analyzes the photos to identify each person's features, such as facial features, hairstyle, clothing style, and so on. Then, it combines these features to create a new image. This image is essentially your face, but may have a movie star's hairstyle and expression, while wearing the clothing from that historical costume photo.

The final result is a completely new photo that displays a unique character: it looks like you, but features the characteristics and style of people from other photos. This is the core function of PhotoMaker—combining the features of multiple different photos to create a completely new, personalized image.

Source code: https://github.com/TencentARC/PhotoMaker