This is one of the most interesting bets that have emerged in recent months
Apple has presented something we didn’t see coming. It is an AI model, something on which it seemed that the company was not betting excessively but which has changed dramatically in recent times. There is already talk that artificial intelligence will be the protagonist of iOS 18, but it seems that the issue is not going to stop there. Apparently, they are also carrying out interesting side projects such as one based on image edition which will allow us to retouch them without having to have prior knowledge of the subject.
In addition to using open source, Apple has published a paper together with the University of California, Santa Barbara in which we can understand in a more balanced way the procedure that has been carried out.
A very interesting project
Apple has made a big splash with AI with the launch of its open source model, MGIE, designed to change the paradigm of photo and image editing by using natural language instructions, similar to what Adobe Firefly does. in Photoshop. This language model allows users without any experience in image editing make complex changes at the pixel levelsimply writing in a text box the modifications you want to make.
Has been designed as Multimodal Large Language Model (MLLM), which gives it the ability to interpret text instructions and apply them directly to images, far exceeding expectations for inference efficiency. This may not mean anything a priori, but the inference is really important. Largely because it is the ability of AI to interpret our language and apply it exactly to what we want. The more inference, the better it will represent our requests in the final image. In something as delicate as editing a pre-existing photograph, it is extremely fundamental.
Unlike other AIs that generate images from descriptions, MGIE is based on the editing of existing photographs. This means that users will forget about editing tools and will simply make text requests to the AI, making the process much simpler and more universal.
- Apple has launched an open source AI model called MGIE to edit photos and images using natural language and without having any idea about photo editing.
- It uses a multimodal large language model (MLLM) to carry out photo editing at the pixel level, achieving great success.
- It has great efficiency in inference.
- Compared to other AIs that imagine the photographs we ask of them, this time it is more of a capacity for edit existing photos.
- It is an open source project and can be seen on GitHub and used if we want. In addition, it has a demo.
- This is something very relevant since it opens a new path to multimodal models to be really useful in some fields of technological knowledge.
We can test MGIE through Github through the following link:
Apple’s MGIE in a Demo Test
To always be up to date with the latest in technology, subscribe to our official and verified Andro4all channel on WhatsApp.