I very much enjoy this feature. My next door neighbor is on vacation, and I'm feeding his fish for him. I took a picture of the fish tank and asked Gemini to put the fish tank at various local tourist attractions in my city, as if we're going on day trips.
I send him one photo a day and he's been loving it. Just a fun little thing to put a smile on his face (and mine).
Fun fact - I trained a lora on our almost-toddler at the time on SDXL and generated images of her doing dangerous things to send to my wife the first day she had a trip away from us.
It was all fun and games until the little shit crawled out of our doggy door for the first and only time when I was going to the bathroom. As I was looking for her I got a notification we were in a tornado warning.
Luckily the dog knew where she had gone and led me to her, having crawled down our (3 step) deck, across our yard, and was standing looking up at the angry clouds.
it can't put two images of people together in one photo, this model still has the issue, also, I have seen cases where Flux Kontext works better in things like removing objects
In quick testing, prompt adherence does appear to be much better for massive prompt and the syntatic sugar does appear to be more effective. And there are other tricks not covered which I suspect may allow more control, but I'm still testing.
Given that generations are at the same price as its competitors, this model will shake things up.