Apple may be last in the AI race, at least when considering competition from companies like OpenAI, Google, and Meta. That doesn’t mean the company doesn’t put effort into technology. In fact, much of what Apple is doing on AI is behind the scenes: While apple intelligence Well, the company’s researchers are working on other ways to improve AI models for everyone, not just Apple users. Latest project? Improve AI image editor based on text prompts.
In a paper published last week, researchers Pico-Banana-400K releasedA dataset of 400,000 “text-guided” images selected to improve AI-based image editing. Apple believes that image datasets improve over existing sets by including a wider variety of high-quality images. Researchers have found that existing data sets either use images generated from AI models or are not diverse enough, which can hinder model improvement efforts.
Interestingly enough, the Pico-Banana-400K is designed to work with: nano bananaThis is Google’s image editing model. Researchers say that Nano Banana allows a dataset to generate 35 different types of edits, as well as Gemini-2.5-Pro to assess the quality of edits and whether these edits should be kept as part of the overall dataset.
Of these 400,000 images, there are 258,000 single edit samples (Apple compares the original images to the edited images). There are 56,000 “preference pairs” that distinguish between failed and successful edit generations and 72,000 “multi-turn sequences” that take 2 to 5 edits.
What do you think so far?
The researchers note that different features have different success rates in this dataset. Global editing and stylization is “easy” and achieves the highest success rate. Object semantics and scene context are “normal”. Precise geometry, layout and typography are “difficult.” The highest performing feature, “Strong Artistic Style Transfer,” which can include changing the style of an image to “Van Gogh” or animated, has a 93% success rate. The lowest-performing feature, “Change the font style or color of text that appears when text is present,” had a 58% success rate. Other features 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 closed to the company’s own platforms, Pico-Banana-400K is open for use by all researchers and AI developers. It’s great to see Apple researchers contributing to public research like this, especially in areas where Apple typically lags. Will we actually get an AI-powered Siri anytime soon?? It’s unclear. But it’s clear that Apple is actively working on AI, perhaps in its own way.
