User captured photos can suffer from a variety of issues such as blurriness, poor lighting, presence of unwanted objects or people, etc. In addition, subjects in the photos can be captured in undesired expressions or poses. This disclosure describes techniques for automatically detecting and fixing problematic aspects of photos by employing generative artificial intelligence (gen AI) techniques. With user permission, a photo can be automatically edited using a suitable generative AI model to fix problems specified in a user-provided prompt. Optionally, a multimodal foundational model can be utilized to detect problems and generate an image correction prompt, along with a prompt that includes the desired corrections. The techniques can be applied to correct a variety of problematic aspects, such as facial expressions, pose, and appearance; lighting; presence of undesired objects or people, etc. Appropriate fairness techniques for machine learning can be incorporated to ensure that the corrections are applied appropriately for diverse photos with different settings and subjects.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.