PhotoMaker allows you to customize realistic human photos through a process known as stacked ID embedding. After providing the AI with an image, you can use simple text prompts to customize it while preserving the key details of the subject. This allows you to do simple modifications like changing the background to changing the clothes the subject is wearing, you can also specify fine details like facial expressions and accessories.
The code is open-source and can be installed on systems with PyTorch and modern GPUs to run demos locally. It’s also possible to run PhotoMaker using ComfyUI and other Stable Diffusion interfaces. There is also a Hugging Face demo available.