tencentarc/photomaker is a personalization generation project that creates images preserving a person's identity based on reference photos of that person. PhotoMaker was introduced at CVPR 2024, and the repository record shows 10090 stars. The provided information lists no recent release.
The core described in the README is the way it personalizes character photos by leveraging Stacked ID Embedding. It explains that customization takes only seconds without additional LoRA training, and it targets ID fidelity and diversity, text control, and generation quality. In other words, it focuses on generating images that match the input prompt while reflecting the character's identity. PhotoMaker is offered as an Adapter for use with other base models and LoRA modules, and PhotoMakerStableDiffusionXLPipeline plus a local Gradio demo are also introduced. The repository's primary language classification is Jupyter Notebook.
The scope centers on personalized portrait photos and stylization generation. The repository's license information is NOASSERTION, and no recent release is recorded either. The provided materials contain no benchmark numbers comparing results with other approaches. Generation quality and the conditions for use are therefore hard to determine from the published description alone; what the materials support is limited to the project's goals and usage paths.