Apple teamed up with the University of California to release the open-source artificial intelligence model MGIE.
Its peculiarity is that it edits photos using text commands. MGIE uses multimodal large language models (MLLM) to understand user commands and simultaneously recognize objects in images at the pixel level.
Artificial intelligence can change contrast, exposure, sharpness, color balance, crop and rotate photos. It can also do complex things, such as editing only certain areas in a photo.
For example, you can say “remove the woman from the background” and MGIE will carefully cut out the person and replace him with a different background.
MGIE can add objects. If you say “make it healthier,” MGIE will add tomatoes and herbs to the pizza photo.
Another example: if you write the command “make the sky bluer”, the AI will increase the saturation of the sky by 20%.
MGIE is available on GitHub, where anyone can find the code and pre-trained models.
You can also test the AI on the Hugging Face Spaces website. MGIE only supports English, and commands currently take several minutes to complete due to the large influx of users on the platform. (VentureBeat)
