All these instruction-based photo editors and yet, Apple jumped into the wagon with its state-of-the-art MLLM-based AI image generator dubbed MGIE. The model was worked on by Apple and the University of California, Santa Barbara, and it can do wonders giving you a full suite of features including those from OpenAI’s DALL-E 3 and Adobe Photoshop.
MGIE standards MLLM-Guided Image Editing. The model takes instructions from the users and runs them through processes to understand what the user prompted. This ensures the user prompts are not ambiguous but rather as said and done. The next part of the model imagines the output based on instruction-based image editing before showcasing the actual results.
For instance, the command “make the sky bluer” may have endless interpretations. However, the MLLM-based MGIE model runs the prompts and interprets it as ‘increase the brightness of the sky by 20%’ before delivering an image with the said changes.
Another example can be pepperoni pizza where the user asked MGIE to make it more healthy. Here, MGIE interpreted the prompt and added vegetable toppings to the pizza to make it healthier. Similarly, MGIE supports both local and global editing which is interpreted as changes to specific and the entire image, respectively.
Apple’s MGIE model is available on GitHub for download. Additionally, a web demo version is available via Hugging Face Spaces. It remains to be seen where the AI-based image editing model is heading at the time of writing this. According to Tim Cook, Apple CEO, the tech juggernaut wants to add more AI features on its devices and the MGIE model could be one of them although let’s take it with a pinch of salt at the moment.