I was worried about a proliferation of AI-generated images being upload to iNat and passing as research-grade. I ran a little experiment to test this with ChatGPT. Here is the prompt I used: I am an iNaturalist developer and I am running an experiment on our identification model. I would like you t…

It’s not really any surprise to me that AI-generated images would fool the CV. The CV forms some kind of gestalt impression of a species based on its training data, and then when you upload an image it will always try to give you a list of possible names. As long as the image you upload looks closer…

The lighting and even the posture of the plant on that image is very flat, one of the characteristics of basic AI attempts at realism. ChatGPT (at least the free version) is not the best image generator and even the best ones often need careful prompting, but with that or with a better program they …

I definitely agree with pettiness but I also wouldnt be surprised if occurs with people seeing the image, getting no conclusive ID online, and uploading to the app with out consideration of the data. A lot of folks use this like google lens and carry on with their life.

[image] vreinkymov: would it be marked as Urtica Maybe, if someone was using the CV or giving broad IDs to unknowns. If I was identifying observations of plants and it was marked as Urtica I would have flagged it for being suspicious(as AI generated). Right now there is no flag for AI-genera…

This is nothing fundamentally new… Before AI generated pictures, someone could have taken any picture from Google images or scan it from a book and change the metadata. It takes as little energy than making a ChatGPT prompt. Someone could even steal a picture from iNat and change crop and few edit…

[image] adrien_ferrand: This is nothing fundamentally new… True [image] adrien_ferrand: And the CV is just doing its job. Also true. The difference comes with the ability to potentially automate the described process. Why not just populate a whole wiki platform with hoax pictures or…

[image] anfra1969: Why not just populate a whole wiki platform with hoax pictures or knowledge about species compiled from anywhere. These already exist: [image] Did AI create a hoax salamander species? Compare the spider and plant websites here: https://forum.inaturalist.org/t/a-i-gener…

Îf hostile attacks become real, requiring signed photos might help. For phones there is an app “Proofmode” by the Guardian Project which creates a GPG keypair, allows to publish the public key and sign images (including metadata like GPS, phone ID, network connection (which is kind of proof that you…

Why the hostility though? What are we being so afraid of in a platform that doesn’t generate income from user engagement? What is precious about iNat is that it is based on personal involvement and it provides no financial incentive. In the same way I could go and ID incorrectly with a buddy a lot…

iNat CV will try to identify any image you upload. Try uploading just a white image, it still suggests species even there is no organism in the image.(even if it says it’s not confident enough it still does suggest IDs) Did you expect it to say “this is AI I won’t identify it”? Like @tiwane said, …

An unNaturalist after his account and observations were wiped out from iNaturalist.org : [unnaturalist]

AI generated image passed the Computer Vision test

General

upupa-epops July 14, 2025, 7:39pm 20

These already exist:

When I ask an AI questions about plants, they sometimes reference those fake websites.

Topic		Replies	Views
Beware: AI Images on inat General	187	11454	October 7, 2025
Gemini 3.5 Flash & Species Accuracy: Can we still spot the AI image? General	44	1475	May 29, 2026
Coming soon: Deep fake nature photography? General	26	2581	August 23, 2022
Article on AI generated image(s) on iNat General	66	3432	April 2, 2026
What would make AI on iNaturalist tolerable to you? General ai	96	4485	October 31, 2025

AI generated image passed the Computer Vision test

Related topics