Apple releases an AI model that can edit images based on text-based commands
MGIE uses multimodal large language models (MLLMs) to interpret your words.
Apple isn't one of the top players in the AI game today, but the company's new open source AI model for image editing shows what it's capable of contributing to the space. The model called MLLM-Guided Image Editing (MGIE), which uses multimodal large language models (MLLMs) to interpret text-based commands when manipulating images. In other words, the tool has the ability to edit photos based on the text the user types in. While it's not the first tool that can do so, "human instructions are sometimes too brief for current methods to capture and follow," the project's paper (PDF) reads.
The company developed MGIE with researchers from the University of California, Santa Barbara. MLLMs have the power to transform simple or ambiguous text prompts into more detailed and clear instructions the photo editor itself can follow. For instance, if a user wants to edit a photo of a pepperoni pizza to "make it more healthy," MLLMs can interpret it as "add vegetable toppings" and edit the photo as such.
In addition to changing making major changes to images, MGIE can also crop, resize and rotate photos, as well as improve its brightness, contrast and color balance, all through text prompts. It can also edit specific areas of a photo and can, for instance, modify the hair, eyes and clothes of a person in it, or remove elements in the background.
As VentureBeat notes, Apple released the model through GitHub, but those interested can also try out a demo that's currently hosted on Hugging Face Spaces. Apple has yet to say whether it plans to use what it learns from this project into a tool or a feature that it can incorporate into any of its products.
