How Apple Plans to Improve AI Image Editors

Apple may be last in the AI ​​race—at least when you consider competition from the likes of OpenAI, Google, and Meta—but that doesn't mean the company isn't working on the technology. In fact, it appears that most of the work Apple is doing is on artificial intelligence. is behind the scenes: Bye Apple Intelligence Well, the company's researchers are working on other ways to improve artificial intelligence models for everyone, not just Apple users. Latest project? Improve AI image editors based on text suggestions.

In a paper published last week, researchers introduced Pico-Banana-400Ka dataset of 400,000 “text-driven” images selected to improve AI-powered image editing. Apple believes its image dataset improves on existing datasets by including higher quality images and more variety: Researchers found that existing datasets either use images generated by artificial intelligence models or are not diverse enough, which could hinder efforts to improve the models.

Oddly enough, Pico-Banana-400K is designed to work with Nano BananaGoogle image editing model. The researchers say that with Nano Banana, their dataset can generate 35 different types of changes, and also connect to Gemini-2.5-Pro ​​to evaluate the quality of the changes and whether those changes should remain part of the overall dataset.

Within these 400,000 images, there are 258,000 single edit samples (where Apple compares the original images to the edited images); 56,000 “preference pairs” that allow you to distinguish between unsuccessful and successful generation of edits; and 72,000 “multi-pass sequences” that go through two to five changes.

The researchers note that different features in this data set had different success rates. Global editing and styling is “easy” allowing for the highest success rates; object semantics and scene context are “moderate”; while the precise geometry, layout and typography are “complicated”. The most effective feature, “strong conveyance of artistic style,” which could include changing the image style to “Van Gogh” or anime, has a success rate of 93%. The worst-performing feature, “change font style or color of visible text, if any,” only worked 58% of the time. Other features tested include “add new text” (67% success rate), “zoom in” (74% success rate), and “add film grain or vintage filter” (91% success rate).

Unlike many Apple products, which are typically closed to the company's own platforms, Pico-Banana-400K is open for use by all AI researchers and developers. It's great to see Apple researchers contributing to open research like this, especially in an area where Apple typically lags behind. Will we get an AI-powered Siri anytime soon?? Not clear. But it's clear that Apple is actively working on AI, perhaps just in its own way.