Apple is joining the AI ​​debates with MGIE, an AI model that manipulates images based on text descriptions. It is still unclear whether and how Apple wants to bring this to its products.

It took a while, but Apple is now also joining in on the AI ​​hype. In a research report, scientists associated with the iPhone company MGIE suggest. This abbreviation stands for: MLLM-guided image editing. It is based on a multimodal LLM that can process text and images simultaneously.

edit pictures

In practice, you can use the model to customize images to your liking. Among other things, you can add, replace, or remove elements from a photo, as well as adjust sharpness and exposure. Some examples Apple gives in the paper include a pizza that is made healthier by artificially topping it with vegetables, or a plate of donuts that turns into a pizza to stay in the same theme. What the model cannot do is create an image itself based on a text description, such as: B. Dall-E 3.

You can test the model yourself on Hugging Face. You have to be patient: the tool quickly becomes overloaded, so waiting times can be up to 6,000 seconds (!). In this phase, MGIE is primarily an experiment with which Apple wants to show that it is also up to date when it comes to generative AI. What Apple plans to do with it in the long term remains to be seen. The Google Pixel 8 and Samsung Galaxy S24 already have extensive AI image editing capabilities, so there is potential for a model like MGIE on the iPhone.