Thanks for the link, this methods sounds workable. The earilest version of MJ results was somehow blurry and noisey, I wonder if it was because of this method.

Vote for Midjourney. I don't know how they improved their performance, no paper or publications.

Sorry I miss understood. You're right.

See the section "ScaNN interface features", you will find that you could search queries with batch, may this similar with your problem?

Maybe you can take a look at ScanNN

Thanks for your long feedback, I've read it twice.

1.re-run the initial query is a great idea, will try to update in the next version.

2.For a ViT-B-32 CLIP model, it will resize all imagines input to the size by 224x224, which is even smaller than that thumbnails, so this will do no harm to performance.

3.Download imagines from iCloud is easy to implement, however it requires network access. It's a disaster for an app that reads all your photos having access to a network, so I made a compromise here.

4.I've tried dynamic scrolling but it cost more time to fetch results, will consider do that way.

5.Search from a few specific album names is a better experience, will definitely find how to implement it.

Really thanks for your patient feedback!

Apple does not allow developer to restrict iPhone model, I'm considering how to block these models to purchase.

It's an completely offline app.

Right.

It's Apple's requirements, I've no choice. Thanks for your advice though

I'm using the simple cosine similarity between embedding vectors. There were some optimized work by Google called ScanNN, which is much faster on large scale vector similarity search. However, it's much more complicated to port this model to iOS.

It's an interesting idea to synchronize the indexes for different devices, however anything related with network connection is a disaster of an app that reads all you photos. Maybe there exists a better way to do this.

On the issue of running on M2, I'll check it out later.

Your project sounds interesting, please get me noticed when there is a product.

Review passed.

I'll check it out, got notice from another user with xsmax not working, I guess it's a chip problem. I'm sorry for that. You can refund first, and I'll also confirm and consider ban the phone before iPhone 11.

Thanks, I agreed. However it's Apple's requirements when I registed Devepleper Account("Fill your Chinese name below").

The calculating speed is of ~2000 photos per minute on iPhone 12 mini.

The time cost for a search also depends on your Photos number, For <10,000 photos it takes less than 1s.

Will certainly check it out!

I am so thankful for your kindness！

It's not on whether it needs permanent nework or not, but on it would request a network access, which will toast pop-up window on first request, which is a privacy and security issue.

Thank you for the suggestion!

No, it's only about content similarly.

In the next version this number will be 120, has submited for review.

I've changed the results number from 12->120, now submitting a new version for review.

Yes, it's a one time permanent purchase.

I agree with you on "free trail then subscription", actually I was going to do the same thing. However, a In-App Purchase requires network connection.

Currently, I'm promoting it at reddit, produchunt, and nowhere, It would be great if you could help me.

Thanks for your useful advice!

1.Great idea！

2.I will change the number to larger(or even configurable) in the next version.

3.Some functions may requires network, but it's cool for the idea of manually adding results.