MGIE by Apple

MGIE by Apple is an open-source multimodal model designed for instruction-guided image editing using natural language. It enables users to perform a wide range of image manipulations by describing the desired change in text, rather than relying on manual editing tools. MGIE supports three primary categories of edits: Photoshop-style adjustments (such as cropping, resizing, color and exposure corrections), global photo optimization (enhancing overall image quality, lighting, and contrast), and local, content-level edits (including object addition, removal, or modification within specific regions of an image).

The model is built to interpret detailed, compositional instructions, allowing multi-step edits to be executed from a single prompt. For example, users can request to “brighten the sky, sharpen the foreground, and remove the person on the left,” and MGIE will attempt to apply all modifications coherently. It is implemented in PyTorch and integrates with Hugging Face and other standard machine learning tooling, making it suitable for research, prototyping, and downstream application development. Use cases include content creation pipelines, rapid visual prototyping, dataset augmentation, and interactive image editing interfaces where non-expert users can achieve complex edits through simple language instructions.