V2 is quite different to V1, with new knowledge bled from uniquely trained sdxl models over several iterations at low weights as not to introduce heavy bias or interfere too much with the Illustrious base. As a result, V2 can produce more realistic images and respond better to natural language whilst maintaining recognition of booru style prompts. It is also more susceptible to negative prompting now and the human female bias has been reduced along with better recognition of different ethnicities, This should make it easies to generate male and trans and even furry/alien characters than V1. In my testing so far it seems to maintain versatility but as expected from a realism model it still cannot match the versatility of Anime Illustrious models. Improved Versatility and character recognition will be the main focus for improvement with V3.
