The artificial intelligence battle is on: there is OpenAI with its ChatGPT with GPT-5 in the oven, Microsoft has its Copilot with GPT-4, Google does the same with Gemini, but what about Apple? Cupertino arrives late to the table, but that is not only not a surprise, but it does not have to be a handicap either, given its imprint in segments such as the smartphone. This week Apple has unveiled MGIE, its first artificial intelligence model.
In the midst of the artificial intelligence furor, Apple has always been cautious and cautious about its plans. Not only that, but while other big tech companies stepped on the accelerator to present their AI tools after the arrival of ChatGPT, Tim Cook’s team remained silent. However, it is known that an Apple GPT-style chatbot exists in Cupertino and important announcements regarding artificial intelligence are expected by the end of 2024. But this MGIE is a very promising start.
What is MGIE and how does it work?
The presentations. A few days ago, a team made up of staff from Apple and the University of California at Santa Barbara released the open source guided image editor MMLM or MGIE, a multimodal model with artificial intelligence capable of editing Photoshop-style imagesonly in this case through simple text commands and AI.
Go ahead, there are already image editors with AI, but as the paper, “human instructions are sometimes too brief for current methods to grasp and follow.” As a result, those touch-ups end up being mediocre or worse than expected. MGIE uses a different approach, using large multimodal language models to understand cues and at the same time, there is image training. Come on, learning from MLLM helps MGIE to better understand natural language commands without especially dense and detailed descriptions.
In some examples listed in the paper, we see how taking a pizza with pepperoni as a base and after adding the command to “make this healthier”, it understands that “this” is the pizza and that it is healthier it interprets as incorporating vegetables. The output is a pepperoni pizza with some vegetables on top. You can also ask her to edit a part of the image, for example by removing the woman from the background or retouching the background of a computer screen.
As a summary: despite being in the first phase of MGIEthis is capable of changing settings related to color, contrast, object manipulation or erasing or, more generally, improving the overall quality of an image.
How you can try MGIE for free and without waiting
Although the MGIE project open source is available on GitHub, if you want to test the model you can do so without having to install anything and from any device (whether from Apple or not), since it uses the browser. You just have to go to this page.
The simplicity of the interface is strikingsomething that considerably speeds up the first steps: it is only necessary to upload an image or drag it to the large box on the left and then add the text instruction in the ‘Instruction’ box.
Finally, you have to tap on ‘Send’ and little else: the result will begin to be generated and displayed in the box on the right. Yes indeed, be patient because the server is limitedwhich implies that if there are many requests in queue, we will have to wait.
Cover | Eva Rodriguez de Luis with Microsoft Designer
In Genbeta | Apple disappoints in artificial intelligence, but it could be the big cover: the reasons to think that it has an ace up its sleeve