Apple has created an artificial intelligence that creates images with just a phrase

Apple has introduced MGIE, a free AI innovation that allows photos to be modified using simple written instructions. This progress was made possible through joint work with the University of California, Santa Barbara.

he apple system It is based on what is called “guided image editing”. This means that using advanced technologies capable of understanding text and images (known as large multimodal language models), MGIE can interpret what a person is requesting and change the image in detail, pixel by pixel.

For exampleif requested by the user Amnesty International “Make the sky bluer” in the picture, MGIE will understand these instructions We will modify the image to meet your request.

The researchers explain that sometimes the instructions people provide are very short and direct, making them difficult to understand with current systems.

However, large language models that understand both text and images (MLLM) show great potential to interpret and respond to these instructions through visual editing.

The developed MGIE method works in two ways with these privileges. First, it takes user-written instructions and translates them into more detailed and specific commands that make editing easier.

For example, if someone wanted to make the grass in an image appear greener, MGIE might understand this as “increasing the green saturation in the grass area by 20%.” This is the obvious command that is used to edit the image.

Then, in a second step, use these advanced models to create an accurate visual idea of ​​what the final edit should look like.

This process creates a guide to adjust the image, adjusting each pixel as necessary. So, MGIE may make specific changes to instructions that are too short or ambiguous Without requiring a lot of additional information and providing results that meet the user's requests.

According to the creators, this system significantly improves the accuracy of edits and user satisfaction, while maintaining a fast and efficient process.

Apple's latest AI model, known as MGIE, has been shared on GitHuballowing public access to the source code, the data used to train the algorithm, the already trained versions, and a guide for use in different editing situations.

This move allows any interested person or developer to explore and try out this technology.

For those who want to try out MGIE without diving into the technical details, an online demo has been live on Hugging Face Spaces, a website that specializes in providing tests of AI tools.

This makes it possible to integrate MGIE into existing applications. Who need to edit photos or create visual content automatically.

This commitment to AI represents a new era in consumer technology, where tools like MGIE not only drive innovation, but also promise to change the way users interact with their devices.

Although MGIE emerged as a research project and Its direct integration into Apple products or services has not yet been announced.This launch confirms the company's interest in strengthening its position in the field of artificial intelligence.

Apple recently unveiled MLX, another algorithm aimed at simplifying the design of machine learning models.

The importance that Apple attaches to artificial intelligence was evident in the statements of Tim Cook, its CEO, who stated that they are working to develop new generative artificial intelligence capabilities that are expected to be released throughout the year.

Experts expect that these innovations will be integrated into iOS 18This represents a major development for Siri, the company's virtual assistant.