How Apple Plans to Improve Its AI-Powered Image Editors
Apple may be last in the artificial intelligence (AI) race—at least when considering competition from companies like OpenAI, Google, and Meta—but that doesn’t mean the company isn’t working on the technology. In fact, much of Apple’s AI work appears to be behind the scenes : while Apple Intelligence does exist, the company’s researchers are working on other ways to improve AI models for everyone, not just Apple users. Their latest project? Improving AI image editors based on text suggestions.
In a paper published last week, researchers presented Pico-Banana-400K , a dataset of 400,000 images with text annotations, curated to improve AI-powered image editing. Apple believes its dataset outperforms existing datasets by including higher-quality and more diverse images. The researchers found that existing datasets either use images generated by AI models or are insufficiently diverse, which could hinder model improvement.
Surprisingly, Pico-Banana-400K is designed to work with Nano Banana , Google’s image editing model. The researchers claim that with Nano Banana, their dataset can generate 35 different types of edits and also connect to Gemini-2.5-Pro to evaluate the quality of edits and determine whether they should be retained in the general dataset.
Among these 400,000 images, there are 258,000 individual edit samples (where Apple compares the original images to the edited images); 56,000 “preference pairs,” which distinguish between unsuccessful and successful generations of edits; and 72,000 “multi-turn sequences,” which span between two and five edits.
The researchers note that different functions had different success rates in this dataset. Global editing and stylization are categorized as “easy” and achieve the highest success rates; object semantics and scene context are categorized as “moderate”; and precise geometry, layout, and typography are categorized as “hard.” The best-performing function, “strongly convey artistic style,” which might include changing the image style to “Van Gogh” or anime, has a success rate of 93%. The least performing function, “change the font style or color of visible text, if present,” was successful only 58% of the time. Other functions tested included “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 only available on the company’s proprietary platforms, the Pico-Banana-400K is open to all AI researchers and developers. It’s great to see Apple researchers contributing to such open research, especially in a field where Apple typically lags behind. Will we see an AI-powered Siri anytime soon ? It’s unclear. But it’s clear that Apple is actively working on AI, perhaps just in its own unique way.