Apple releases ‘MGIE’, a revolutionary AI model for instruction-based image editing


Apple has released a new open-source AI model, called “MGIE,” that can edit images based on natural language instructions. MGIE, which stands for MLLM-Guided Image Editing, leverages multimodal large language models (MLLMs) to interpret user commands and perform pixel-level manipulations. The model can handle various editing aspects, such as Photoshop-style modification, global photo optimization, and local editing.

MGIE is the result of a collaboration between Apple and researchers from the University of California, Santa Barbara. The model was presented in a paper accepted at the International Conference on Learning Representations (ICLR) 2024, one of the top venues for AI research. The paper demonstrates the effectiveness of MGIE in improving automatic metrics and human evaluation, all while maintaining competitive inference efficiency.

How does MGIE work?

MGIE is based on the idea of using MLLMs, which are powerful AI models that can process both text and images, to enhance instruction-based image editing. MLLMs have shown remarkable capabilities in cross-modal understanding and visual-aware response generation, but they have not been widely applied to image editing tasks.

Screenshot 2024 02 06 at 1.02.29%E2%80%AFPM
credit: arxiv.org

MGIE integrates MLLMs into the image editing process in two ways: First, it uses MLLMs to derive expressive instructions from user input. These instructions are concise and clear, and provide explicit guidance for the editing process. For example, given the input “make the sky more blue”, MGIE can produce the instruction “increase the saturation of the sky region by 20%”.

VB Event

The AI Impact Tour – NYC

We’ll be in New York on February 29 in partnership with Microsoft to discuss how to balance risks and rewards of AI applications. Request an invite to the exclusive event below.

 

Request an invite

Second, it uses MLLMs to generate visual imagination, which is a latent representation of the desired edit. This representation captures the essence of the edit and can be used to guide the pixel-level manipulation. MGIE uses a novel end-to-end training scheme that jointly optimizes the instruction derivation, visual imagination, and image editing modules.

What can MGIE do?

MGIE can handle a wide range of editing scenarios, from simple color adjustments to complex object manipulations. The model can also perform global and local edits, depending on the user’s preference. Some of the features and capabilities of MGIE are:

  • Expressive instruction-based editing: MGIE can produce concise and clear instructions that guide the editing process effectively. This not only improves the quality of the edits but also enhances the overall user experience.
  • Photoshop-style modification: MGIE can perform common Photoshop-style edits, such as cropping, resizing, rotating, flipping, and adding filters. The model can also apply more advanced edits, such as changing the background, adding or removing objects, and blending images.
  • Global photo optimization: MGIE can optimize the overall quality of a photo, such as brightness, contrast, sharpness, and color balance. The model can also apply artistic effects, such as sketching, painting, and cartooning.
  • Local editing: MGIE can edit specific regions or objects in an image, such as faces, eyes, hair, clothes, and accessories. The model can also modify the attributes of these regions or objects, such as shape, size, color, texture, and style.

How to use MGIE?

MGIE is available as an open-source project on GitHub, where users can find the code, data, and pretrained models. The project also provides a demo notebook that shows how to use MGIE for various editing tasks. Users can also try out MGIE online through a web demo hosted on Hugging Face Spaces, a platform for sharing and collaborating on machine learning projects.

MGIE is designed to be easy to use and flexible to customize. Users can simply provide natural language instructions to edit images, and MGIE will generate the edited images along with the derived instructions. Users can also provide feedback to MGIE to refine the edits or request different edits. MGIE can also be integrated with other applications or platforms that require image editing functionality.

Why is MGIE so important?

MGIE is a breakthrough in the field of instruction-based image editing, which is a challenging and important task for both AI and human creativity. MGIE demonstrates the potential of using MLLMs to enhance image editing, and opens up new possibilities for cross-modal interaction and communication.

MGIE is not only a research achievement, but also a practical and useful tool for various scenarios. MGIE can help users create, modify, and optimize images for personal or professional purposes, such as social media, e-commerce, education, entertainment, and art. MGIE can also empower users to express their ideas and emotions through images, and inspire them to explore their creativity.

For Apple, MGIE also highlights the company’s growing prowess in AI research and development. The consumer tech giant has rapidly expanded its machine learning capabilities in recent years, with MGIE being perhaps its most impressive demonstration yet of how AI can enhance everyday creative tasks.

While MGIE represents a major breakthrough, experts say there is still plenty of work ahead to improve multimodal AI systems. But the pace of progress in this field is accelerating quickly. If the hype around MGIE’s release is any indication, this type of assistive AI may soon become an indispensable creative sidekick.

VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.



Source link

About The Author

Scroll to Top