Apple open-sources an AI tool that edits photos with plain text

MGIE is an open-source image editing model developed by Apple in collaboration with researchers from the University of California, Santa Barbara. Short for MLLM-Guided Image Editing, it lets you edit pictures entirely through plain text without ever opening professional photo editing software.

What it does

Beyond basic adjustments like cropping, resizing, flipping, or adding filters, MGIE really shines when handling complex, context-heavy instructions.

Instead of fiddling with adjustment layers, you simply type what you want changed. In the research paper, the team showed an image of a pepperoni pizza; when given the prompt "make it more healthy," MGIE automatically topped the pizza with vegetables.

In another test with a dark photo of tigers in the Sahara, typing "add more contrast to simulate more light" made the scene noticeably brighter while preserving natural textures.

How it works

MGIE combines two distinct superpowers of multimodal large language models.

First, it parses ambiguous user language into explicit visual intentions. If you say you want a bluer sky, the model figures out that it needs to bump up brightness and saturation specifically on the sky portion of the image.

Second, it applies those calculated modifications directly to the visual pixels, generating reasonable edits while preserving the rest of the composition. The researchers noted that this framework significantly boosts editing accuracy while keeping processing efficiency competitive.

Where to try it

Apple has made the code for MGIE available to download directly on GitHub.

For those who prefer not to run code locally, VentureBeat reported that a public web demo is also hosted on Hugging Face Spaces, where you can upload an image and test out text prompts right away.

Apple hasn't announced any commercial roadmap for the model yet, keeping it strictly in the realm of open research for now.

Why it matters

While Adobe uses Firefly to power generative fill in Photoshop and OpenAI lets you tweak DALL-E 3 images with prompts, MGIE shows what natural language editing looks like when applied to your own existing photos. It cuts out the technical jargon of photo editing, letting anyone direct visual changes by simply describing the outcome.

We think this gives us a fascinating preview of where Apple's mobile software could go. Even though MGIE is currently just a research prototype with rough edges around speed and complex scenes, it is an essential project for developers exploring text-guided computer vision to test right now.