Back in December of 2022 and then in April of 2023, Midjourney released their Niji models v4 and v5, respectively. The Niji model specializes in anime and illustrative styles. I never took the time to explore it, so I'm going to hop right into Niji V5.
Here's a view into the 5 styles that we can choose from using the prompt
a spear warrior in unusual armor, action shot --seed 286950974 --s 250 --niji 5 in high variation mode with each image's title as the argument for the
Default, Scenic, and Original feel very much like highly-detailed concept or cover art, with Scenic feeling the most action-focused. Cute definitely gives off chibi vibes. Expressive looks pretty close to the art style of a lot of manhwa (Korean comics), with blocks of colors and lots of fluff and layers and relatively less texture.
The docs state that the model is sensitive to the
--s stylize parameter. How does stylize affect the output? Here's the same prompt in the scenic style with different levels of stylize.
I don't get a solid feeling for how this parameter affects the output. Perhaps the backgrounds get a little bit more complex and detailed? I find it interesting that the bottom two characters switched places between 50 and 100 and gained something like a bow between 100 and 250.
How do different models compare?
Let's look at the 6 latest models with the same prompt and seed. Each of these have been generated with
offroad dirtbikes in the desert --seed 1773537108 --s 100 <model> upscaled 4x if possible, in high variation mode.
There's a significant change in the Niji V4 and Niji V5 models, with V5 being vastly more illustrative. There same improvement is seen in the regular V4 and V5 models. With that, I won't bother exploring Niji V4.
With all of that aside, let's see what Niji V5 can make across all of its styles. Enjoy!