Hello, thank you for your answers! I already tried to profile various iOS devices, without success: from iPhone 12 to 16 Pro: all are only capable of 2 requests in parallel. Requests must definitely be .accurate, and resizing the image would be too resource demanding. I'm flying blind here unfortunately, and currently, I have to serialize and avoid having more than 2 in parallel, otherwise the app completely hangs. I wish I could at least have some error thrown.

Thanks for the lead. Actually, I was very happy to use RecognizeTextRequest.perform(on: URL) because I was supposing that the underlying code was optimizing everything in terms of CPU/GPU. Of course it wouldn't make any sense that the code would resize the image there, not knowing what the image contains. Therefore, If I'm to resize the images before OCR by defining a maximum resolution, do you have a recommandation (ballpark) on good pixel values to use knowing that the images are documents? Also, can you confirm that RecognizeTextRequest.perform(on: CIImage) would be the optimized choice to use in the case of a resized image? In other words, flow would be: URL > CIImage > resized CIImage > OCR Thank you very much in advance :-)

I have this issue as well.

It's in the top-left of the screen. multiply.circle button

It's a workaround, but why is it happening?

FB19946456

See Apple doc: https://developer.apple.com/documentation/foundationmodels/generating-content-and-performing-tasks-with-foundation-models

This is not the same as my issue. Content is correctly indexed since a query will return it. But Spotlight itself will never return the object if the actual string within textContent is searched. Now that I write this, I might get what you mean. Maybe Spotlight won't find it because it queries textContent like you.