Leafhoppers vs AI - checking on progress

I'm currently writing a short article on Ribautiana species, so naturally rather than get on with it, I took the opportunity to be distracted by playing with AI. Recently Google released an update to its Gemini AI engine and this includes improvements to its (annoyingly named) image generator nano banana. Having spent a long time staring at photographs of Ribautiana and having zero artistic ability, I thought it would be useful to see if Gemini was capable of generating a labelled summary diagram of the differences between the UK species. You can see the result above. It didn't quite work but there is considerable promise.

First the easy part - the things it got wrong brackets (everybody loves to criticise AI):

• It didn't get the position of the dark dots on R. ulmi right, although it did get the idea that those dots are important for identification. 
• It made a mess of the apical dark marks on the wings of R. tenerrima.
• It made a complete mess of the aedeagi drawings. 

What it got right:

• It picked up on the fact that only two species of Ribautiana are visually distinguishable, at least in some cases, while the other three species generally are not. 
• It got the idea that the dark dots on R. ulmi are important and that the pale lines on the forewings of R. tenerrima are also significant.
• Most importantly, it correctly confirmed that dissection is important for definite identification of these species.

I didn't spend any time refining the prompts and getting Gemini to redraw the image - I suspect that if I had, the result would have been pretty good. I was really interested to see how AI is progressing and it certainly shows promise. And now, if you're suffering from lack of reality, here's the real thing: