I found a new AI tool Shakker, a best image to image tool. You can try it via https://www.shakker.ai
The production of this Lora actually originated from helping a friend to take pictures of seawater corals, but the available models found were not particularly suitable.
The training source uses some real pictures and some midj5.2 images, 768*768 is the original training size
The main trigger word is "underwater" and "close up" can be used to increase the close-up draw rate. The weight is recommended to be 0.6-0.9
There will be some nice light and shadow, and then there is a high probability of coral (possibly strange succulents) and stones
It can have a good performance in 2D-real style (it can cooperate with my other Lora to produce a real sea view chassis)