PhotoMaker
An AI portrait generator you can run locally, using reference images and text prompts to preserve identity without extra LoRA training.
PhotoMaker generates realistic portraits and stylized images of a person from reference pictures and text prompts. You can run it locally through a Gradio interface, including on macOS, with community implementations for Windows. It's aimed at people creating personalized photos, artwork or avatars who want to retain a person's recognizable features across different scenes.
It doesn't require additional LoRA training for each person. Instead, it combines the identity information from several reference images into a shared representation. That approach lets you supply more pictures of the same person to improve resemblance while using text to describe the setting and appearance.
The references can include paintings, sculptures and old photographs, so the input doesn't have to be a modern portrait. PhotoMaker can render those subjects as realistic people, apply artistic styles, or change age and gender while retaining identity attributes. It can also blend features from different people into a new identity, with control over each person's contribution.
For users who already work with image generation models, PhotoMaker acts as an adapter alongside other base models and LoRA modules. Community integrations include ComfyUI and WebUI. The local implementation uses Python and PyTorch; hosted demos on Hugging Face and Replicate provide a separate way to try generation through cloud services.
