I agree, it still takes vigliance and awareness on the community’s part, so definitely something to be on the lookout for. If something is AI generated, please mark “No” for “Evidence of organism” in the Data Quality Assessment.
I’m not sure what their reasoning was, but that’s not the case. Use whatever aspect ratio you like.
To be fair, there are multiple places where square photos look best, including taxon photos, thumbnails, and the identify portal. Other aspect ratios work, but don’t look as good at default zoom.
I’ve heard that too. I think it’s mainly because thumbnails and icons used on the site are square, so non-square photos sometimes crop poorly in those views, but another non-issue as far as I’m concerned.
The toad looks like it can be 3D rendered. 3d Objects of animals can be obtained, a skin added to it. The software simulates light on the object. The simulated light effects make the creature realistic. Fur is a very tricky to add to 3D models compared to smooth surfaces, but the software today might be able to render that too. The computer speed is amazing these days. or maybe there is another way to get simulated pictures without going through the whole 3D route.
It is a fake impression some entities are trying to create of how powerful AI is. It draws in capital funding, investments, creating randomised anomalies. That’s what makes people curious to get into it.
If you’ve got a long vine or a tall mushroom or a tree, the aspect ratio is better as a rectangle.
I used craiyon, because most of the others you need to sign up for with an email address
If it comes to the worst, there is a way out: digitally (GPG-) signed images combined with signed phone sensor data, all nicely packaged into a zip file. The Guardian Project had an app for that:
As far as I understand, there come two signatures with each image: one locally generated (the app creates a GPG keypair during installation) and one (optional) offline made by their cloud server.
This is intended for journalists and activists to be able to prove that they really took the picture at the time and place they claimed, e.g. to document police violence. It would require significant additions to inat’s infrastructure to check those kind of data, though.
PS: here is a screenshot of the file manager after taking a pic of a jar of bruschetta. The first line is the zip file generated by the app, the rest are files after manually unzipping that.
That toad looks like Trump in his mugshot!
I always crop my photos to square, since that is how they are usually displayed, and the automatic crop might not be so good.
What immediately struck me about that 2nd fox photo is the depth of field. The foreground and background are blurred out (as one might expect), but the entirety of the animal, in a face-on, longitudinal view, is in crisp focus. It seems that would take some extraordinary circumstances and/or fancy camera settings to accomplish.
Agreed. What really concerns me is when photo fakers start instructing the AI to “make it a little out of focus with somewhat poor contrast and depth of field so it looks more like a photo an amateur would take.”
AI does not generally understand light sources at this point.
Some interesting work done on detecting faked digital photos over 15 years ago was looking at exactly this. It takes sophistication to manually fake an image and get the light and shadows correct, and it may be extremely difficult to do it in a way that software can’t flag.
It’s only a matter of time until generative AI learns/is taught to be even more realistic.
Seconding this note of caution. I asked someone the other day on a fox picture I was initially convinced was AI, but they replied it was real. I think some pics look “off” because of the post-image processing that some phone camera apps do, or the distortion created by digital zoom.
I’m afraid it might soon if enough people ask for it. Just instead of generating an image, let it generate a source file for some raytracer or whatever rendering engine. It will be a little harder to create something that looks like the object to be displayed, but the lighting/reflections/shadows will be automatically correct.
Waiting for the day someone uploads one of the AI generated " Opium Birds" or one of the other many AI memes and posts that are sweeping the internet as an Inat observation as is often the case with certain users posting random memes onto Inaturalist for no apparent reason. maybe that’ll be my motivation to check the human obs for anything of that sort ( and also flag it for copyright?)
That’s already happened, some kid posted an “opium bird” the other day and I ended up having to flag it.
I cited this thread in an article for my local mycological society on why image identification apps are not particularly useful for fungi. (Context: there are lots of data quality issues, and AI-generated images are one of them).
How useful are phone apps for identifying fungi?
But iNat prefers to use the term Computer Vision - they say it is not AI.