Many contemporary art styles have come and gone, but very few survived, and even fewer thrived with age. Anime is an anomaly. It's been a huge part of Japan's culture for the last century and has since become one of its largest industries.
AI image generators may have been amazing, but they couldn't mimic the intricacy and passion behind talented animators. Midjourney aims to capture that magic with its Niji model, and it's possible that they've already done it with Niji 6.
The only question is: how good are we talking?
Throughout this article, I'll be putting Niji through its paces with a variety of prompts based on different genres of anime. We'll explore its strengths and weaknesses, share a couple of tips for getting the most out of the model, compare it with its previous model and base V6, and showcase some of the amazing creations that only Niji can produce.
What is the Niji Model?
Niji is an AI art generation model made by Midjourney in collaboration with Spellbrush, an AI generative studio that specializes in anime art. This newest model borrows heavily from Midjourney V6’s latest features, allowing it to understand prompts better and create understandable (but imperfect) text.
How Can You Use the Midjourney Niji V6 Model?
If you’ve used Midjourney in the past, then using Niji will be a walk in the park. You just need to attach the “--niji 6” parameter at the end of each prompt. Since it’s a strong model, you might also need to turn down the stylization to 250 or lower. For example, here’s an image of Luffy from One Piece at “--s 250” and you can already see that it has a stronger style even at 250.
Some Examples of Midjourney Niji V6
I’m not going to tease you any longer.
Here are some examples of Midjourney Niji V6, along with the prompts that created them:
Base Prompt: an astronaut landing on an advanced moon base
Base Prompt: lush alien forest at night, bioluminescent plants, floating jellyfish in the sky
Base Prompt: a technologically advanced city built on the underside of a giant asteroid, inspired by cowboy bepop
Base Prompt: eren from attack on titan
Base Prompt: a friendly ghost in the middle of a field, inspired by spirited away, studio ghibli, hayao miyazaki
Base Prompt: a woman protagonist who can control monsters based on the four basic elements, getting ready for a battle, epic anime scene
Base Prompt: a creepy dimly-lit hallway, inspired by corpse party 2013
Base Prompt: a man watching an unknown spiral in the sky, unfathomable, junji ito
Base Prompt: a woman with blood on her face, inspired by perfect blue 1997, satoshi kon aesthetic, analog
Base Prompt: A colossal and battle-scarred mecha emerges from the depths of a volcano, mappa animation
Base Prompt: Two sleek and futuristic mechas engage in a high-speed aerial dogfight, laser beams, interfacing
Base Prompt: A mecha and kaiju hybrid with electricity powers, mappa, rain, atmospheric, evening
Base Prompt: exterior, a coffee shop named "Kohi"
Base Prompt: yui hirasawa holding a sign that says "Good Morning", in the style of K-On!
Base Prompt: a mage casting a rune, fantasy
Base Prompt: a beautiful blonde girl with one green eye and one brown eye, heterochromia, slice of life
Base Prompt: a woman with her basilisk, fighting stance, character design
Thoughts So Far
These are absolutely incredible. The prompts are more accurate, it has a deeper anime database so it knows who I'm talking about when I specify a character, you can do text, and it's expressive.
However, if I were to be nitpicky, I’d say that Midjourney really understated the strength of this new model. Even at --s 100, which should only be a tenth of its stylization power, it’s still really opinionated. My only hope is that they’d recalibrate the model in the future for a more accurate generation.
Direct Comparisons With The Niji 5 and the Base Model
In this section, I’m not going to generate the base model images with the same prompt. Instead, I’ll attach an “anime art” or something similar at the end of each base model for a better 1:1 comparison.
Thoughts So Far
Quality-wise, I’d say that Midjourney V6 and Niji 6 are pretty close. However, Niji 6 is more faithful to creating anime art, while V6 is broader and more flexible with art (as it should be). On the other hand, Niji 5 was great for its time, but after seeing them side by side, it’s pretty clear how big of a difference there is between these two Niji models.
This newest Niji 6 model is another great addition to Midjourney's already capable arsenal. While not perfect, Midjourney showed an ability to create anime characters, backgrounds, and scenes when prompted with the right inputs. The images are incredibly rich, faithful to the source material, and imaginative.
They do, however, have to improve their stylized parameters to be more accurate. But once they have that sorted, I struggle to see other areas of improvement for this model besides getting better at imitating certain styles.
Midjourney's Niji 6 model is amazing for anime artists, creators, and even just fans of anime in general. If you're not satisfied, I highly suggest reading our other articles on Midjourney. Have fun!