Apple has publicly released AI research that involves an open-source model capable of image editing based on natural language commands. This model, called “MGIE” (MLLM-Guided Image Editing), was developed and researched by a team from Apple and researchers from the University of California, Santa Barbara. The MGIE model combines a large-scale language model with image data to provide various ways of editing images, ranging from pixel-level modifications to overall image editing.
In the research presentation, examples were shown to demonstrate the model’s ability to interpret commands and edit images to closely match desired outcomes. For instance, when instructed to make a pizza image healthier, the model added vegetables to the image. It can also perform detailed edits, such as removing a person from the background or modifying a computer screen in the image.
MGIE’s capabilities include overall image ambiance editing and Photoshop-like feature adjustments such as cropping, resizing, rotating, adjusting lighting, sharpness, and brightness. It can also edit specific objects identified in the image.
MGIE is an open-source project that can be further explored on GitHub.
TLDR: Apple has published research on an open-source AI model called MGIE, which allows image editing based on natural language commands. The model offers various editing capabilities and can interpret detailed instructions to modify images. MGIE is available on GitHub.
Leave a Comment