I ask this here to see whether my perceptions are mine alone. Over the last couple of weeks I started using the inat iphone app and it’s pretty neato. It’s great to click the + sign, take a few photos, click Identify and save. When I go back to sit at my pc to go over the photos, photo quality is surprisingly poor. Why is the quality level so low? There’s plenty of bandwidth. I guess I’ll just have to go back to taking photos and adding them manually which doesn’t have that effect.
Do you mean that if you take photos with your iphone and upload them via the website, the photos are better than if you take the same photos on the iphone and upload via the iphone app?
You could try taking the photos first and then uploading them via the app.
Phone camera apps use A LOT of computational processing to get decent images from such tiny lenses and sensors. Modern phones use a lot of AI-powered processing automatically.
https://forum.inaturalist.org/t/oversaturated-unnatural-iphone-color/31938/21?page=2
https://applemagazine.com/iphone-camera-processing/
It’s my understanding that the iNat app (as well as other non-native camea apps) can’t leverage all of those processes, so the image quality appears less good. But there’s been discussion on whether the AI-powered processing is welcome or not…or to what extent it may be “making up” detail that doesn’t exist or smoothing out detail that does.
It would be nice if the iNat staff could provide some details on the difference between the native camera apps and the iNat app.
Here’s a side-by-side comparison of stock Samsung Galaxy S24 camera app and photo taken with the Android iNat app (original and cropped to show detail).
This is most noticeable to me in closeups. I need to experiment with using magnifying outside of the camera to see if that helps.
So are you using the multi-photo camera, or the camera with in-app identifications?
I’m not sure this is accurate, and “AI” is a loaded and broad term. My understanding is that most phones don’t use generative AI by default, as in they’re not making up content, but they may use AI to examine the photo and decide on exposure levels, blending frames, etc. But yes, they use a lot of post processing to try and get a usable photo from a tiny sensor and to allow people to take photos in all kinds of situations without any special skills or knowledge.
I’m using an old trusty and crusty iPhone 11, OS 26 installed, with iNaturalist 1.2.0 (234).
Inner thoughts: It’s very good at a specific thing which is a wonderful thing for the community to have, which is opening an app, seeing an unambiguous button to press, pointing the camera, taking the picture, and saving the observation. You do not need to wait for geocoords to stabilize, nor even identify, though you should. The app is great for bioblitzes and just going bonkers with in general. I’m teaching my kids using the app, which they love. It’s shallow learning, but it is learning, and a gateway to deeper understanding at a later time. Ok then.
The photos iPhones are capable of capturing are HEIC, which effectively “stacks” short video bursts (MOV) into a single image, which is processed by the camera app. You can get even fancier with other software to review the frames in the MOV and delete poor ones to improve the HEIC, which you can then export as a PNG / JPG etc. I don’t ever feel compelled to get fancy. I click the button, out comes a photo. Happy. There is nothing AI about it. It is native to the OS.
Why HEIC is good? It is faster to capture video. When you want to look at a single frame, then it processes the video to provide a composite of the “best” parts. This helps identify camera shake and reduce blur. Faster shutter speed = better quality for HEIC.
I had my camera settings at default, which uses the HEIC technique. I expected iNaturalist to use the settings I had made. The images that the inaturalist app captured were inferior. Why is iNat not using my settings.
…is it because it wants to shrink an image down to barest minimum for real-time identify? In doing so, so much image quality is lost and this impairs the identify function as to wreck its quality. Is this the cause of the image degradation because those are then uploaded, rather than the original higher quality photos?
I take photos
The iNaturalist computer vision model used to be referred to as machine learning (and we were corrected when users called it AI). Now iNat staff refers to it as AI. Google defines AI as including ML.
https://cloud.google.com/learn/artificial-intelligence-vs-machine-learning
AI, as defined by Google, is used extensively in phone cameras as the default, automatic processing–beyond determining exposure, etc. The jury is still out on how much of it could be accurately be labeled generative AI–it’s minimal for sure (currently).
https://www.bgr.com/2135558/smartphone-uses-more-ai-than-you-think/
This thread keeps refering to smartphone use of AI in processing photos as if it is one “thing”. Given the multiplicity of phones, cameras, and especially the settings available to each user, no generalizations across devices/platforms/apps are likely to be very useful.
As I’ve mentioned previously, “I absolutely hate all sweeping generalizations.” 
Correct. Each device, and even software updates on the same device, will employ different computational processing and to different degrees. The generality is that the stock cameras on any device will produce a different photo (often better, but not always or in every way) than a photo app that is not stock–because the stock camera app will have greater access to the AI/ML computational processing than non-stock photo apps). That’s the focus of the discussion (stock vs. not stock photo apps like the iNat app).
HEIC isnt a stack of video bursts. It’s just a compressed image format similar to JPEG: https://en.wikipedia.org/wiki/High_Efficiency_Image_File_Format
Most stock camera apps nowadays take a burst of images and combine them in various ways to create an optimum image using various types of computational processing (usually involving AI/machine learning).
The iPhone 11 was actually a key moment for Apple’s AI-powered computational photography. Key AI features were introduced for the first time. These included use of the A13 Bionic chip’s Neural Engine for “advanced machine learning” to combine 9 images into one optimized shot, optimizing texture, details, and noise. the iPhone 11 also uses Smart HDR (computational photography that combines multiple exposures using AI).
Based on what I’ve been reading, Apple’s Deep Fusion and other advanced AI features require specific integration with the camera chip’s Neural Engine that is primarily reserved for the native camera app. Third-party (non-stock) apps can evidently access some baseline image processing through iOS, but they evidently don’t get the full multi-frame analysis and per-pixel optimization that the stock app does.
From what I’ve read, stock camera apps typically do not use true generative AI automatically during capture. They evidently use traditional computational photography AI, which is different from generative AI. But it’s rather unclear where the line between the two is (hence “typically” and “evidently”).
But again, this thread is about comparing photo quality between stock and third-party (non-stock) apps…assuming that’s what the OP is referring to (which is unclear).
I bow to your out-nerding me. I’m glad people know what they are talking about here.