We are not yet at the point where artificial intelligence floods our iPhones, which has already been confirmed for later and will go hand in hand with iOS 18. However, we can consider a small preview of what Apple presents now with ‘MGIE’, a new AI model capable of editing existing images using Photoshop style prompts (different from creating tools like Midjourney or DALL-E 3 from scratch).
This is a open source project which was developed by Apple itself and by the University of California at Santa Barbara. Although we are in the first phase, the truth is that it already offers good results and anyone can try it.
What is MGIE and what does this artificial intelligence allow?
MGIE is an MLLM, which if you don’t know its meaning, is the English acronym for large-scale multimodal language models. It combines with the so-called broadcast model and allows simple instructions sent by text to be transformed into concise instructions resulting in an image transformation.
And what does all this translate into? Well, by obtaining functions similar to those integrated by Photoshop’s generative AI, although in an even simpler way. We can see this with the simple examples provided by the University itself and Apple, in which a simple instruction is shown by the user and the transformation of this into a much more precise text and, after that, modifying the image.
Example: The user stands in front of an image of a spicy salami pizza and says “make it healthier”, and the AI interprets the text and image to automatically get clearer instructions on how to make this healthier pizza. For this you get something like “pizza includes vegetable toppings like tomatoes and herbs”. AND SoAfter that, the pizza image now shows, in addition to salami, much healthier toppings.
Examples are also presented in which this AI is capable of edit only part of the image. Consider asking for the sky in a photograph to be bluer and increasing the saturation by 20% when editing. Or that content disappears on a computer screen and turns completely green.
Because yes, among all the capabilities added to MGIE there is the possibility of change color and contrast settings, manipulate objects, delete them, etc. Generally speaking, this can also be used to improve the quality of an image when it is very saturated or blurry.
You can now access the trial version of MGIE
The entire repository with the MGIE open source can be found on GitHub, although if you want to access a test of this model now there is the possibility of doing so through this page. Needless to say, you can try it from a Mac, iPhone or iPad or from any other device, since it is through the browser.
There you will find a very simple interface in which you just need to look left at the start, in which you have to upload the image from your computer or mobile, add simple text instructions from ‘Instruction’ and finally click on ‘Send’.
Behind that the result will start to be generated and you can see what instruction was automatically given to the final image and example. Of course, it should also be warned that the process could be delayed, because the server is limited and when there are many pending requests, ours remains in the queue.
Will we see something like this natively on iOS and macOS?
It was already significant to see the rumors about Apple and AI for 2024 and that Tim Cook himself hinted that they were preparing something important for this year. And although this responds to a local university project, the truth is that it serves at least to know that Apple is already exploring the field of image generative AIapart from the text.
It is obvious that this confirms absolutely nothing, but at least it makes us dream. If Apple is able to develop a project like this as open source, who knows if they already have a tool in-house to improve image editing on iPhone, iPad and Mac. In June we will dispel doubts.
By | Xataka Mexico
More information | University of California and Apple
Cover image | Álvaro García M. with DALL-E 3
In Applesfera | All the Samsung Galaxy S24 Ultra AI I want on my iPhone and where Apple stands
In Applesfera | Meet Pi, the new artificial intelligence that triumphs in WhatsApp iOS