The technological giants do not slow down in their race to be at the forefront of the development of Artificial Intelligence. OpenAI works on Microscope, the neural analysis AI; Google has renamed Bard “Google Gemini” and, in this case, Apple has just launched MGIE, a new AI model that allows you to edit images using natural language.
How MGIE works, Apple's AI for editing photos
The Cupertino company and the University of California at Santa Barbara (UCSB) have joined forces to create a new open source model that is capable of edit images based on natural language instructions.
Despite the existence of a multitude of tools that are responsible for creating images, they have focused on “guided image editing” taking advantage of large-scale multimodal language models (MLLM), which are those learning models that are trained with large amounts of data.
How to test MGIE
At the moment it is a research project, so it is not yet known for sure whether Apple will end up incorporating this type of AI into its publicly available software. What is clear is that they are going to continue researching this type of tools.
In any case, MGIE is available as an open source project on GitHub, where users can find the code, data, and pre-trained models. Furthermore, it can be tested by online trial version which Hugging Face Spaces makes available, through: https://huggingface.co/spaces/tsujuifu/ml-mgie
Of course, it is necessary to test it using the English language, the only one available at the moment.
What MGIE can do
What MGIE does is interpret the image that is offered and the order that is given in the form of a prompt to proceed with the editing. Thanks to the fact that it consists of an MLLM, it is able to derive concise expressive instructions and give explicit visual guidance.
In fact, one of the goals of Apple and UCSB researchers is to demonstrate that this type of language model can help make image editing with AI easier.
Broadly speaking, what this AI can do is Photoshop-like tasksFor example, some simple ones such as making simple color and lighting adjustments, but also more complex ones such as removing an object from the background, changing the background or even editing a specific area of the image without modifying the entire image, so manages to make local editions.
Let's understand it with some examples. Given the image of a pizza and the order to “make it healthier,” the tool takes the concept and what it does is add vegetable ingredients, seeking to resemble human expectations.
On the other hand, if a photograph is entered and the prompt “make the sky bluer” is sent, the AI will interpret it and what it will do is increase the color saturation, but only in the areas where the sky appears.