New And Improved Niji V6 Image Model Is Here

It’s been a while since Midjourney last dropped an update on its AI models. Most groundbreaking news in the AI world lately has been about AI videos and language models. But a few days ago, Midjourney released a minor upgrade to its anime-style image model, Niji. This update, while not revolutionary, brings some welcome improvements to the table.

For those who aren’t familiar with Niji yet, it’s Midjourney’s model specifically designed for Eastern and anime aesthetics. I am not much of an anime enthusiast, but I’ve been following Niji’s development since 2022.

What’s new in Niji 6?

The new updates primarily focus on text rendering and better image quality. Take a look at some examples the Midjourney team shared on Discord.

New model updates:

• Improved Japanese text rendering: This is perfect for users who want to incorporate authentic-looking Japanese texts into their images.
• Can also render Chinese characters: The model not only renders short Japanese kana text well but, in some cases, can also render text that includes simple Chinese characters.
• Improvements in image details: One notable enhancement is the structure of anime eyes. As any anime artist knows, eyes are crucial in conveying emotion and character, so this improvement is particularly exciting.
• Fixes some image artifacts that previously affected a very small number of generations.

How to Enable Niji Model V6

As of now, there’s no way to switch to the Niji model from the MidJourney Alpha, and features like inpainting, panning, and zooming will be available at the full launch in approximately a month. So, in order to use the new model, we’ll have to use the Discord chabot.

Toggle the settings menu by typing “/settings” and pressing the enter key. In the model list, select “Niji Model V6 [ALPHA]”.

Once you’ve done this, you’re ready to start generating anime images.

Examples

Here are some examples I created with Niji model V6.

Prompt: girl with white hair, one-eye closed, holding a sign that says “愛してる”

Here’s an upscaled version of the fourth variation:

The results were impressive. The model accurately rendered a girl with white hair, one eye closed, holding a sign with the Japanese text “愛してる” (which means “I love you”). The text was clear and legible, demonstrating the improved text rendering capabilities.

However, it’s not always accurate. When I asked Midjourney to render an image of a ship with a Japanese label “ありがとう” (meaning “thank you”), I encountered some issues.

Prompt: a ship with a label “ありがとう” — ar 16:9 — s 250 — niji 6

None of the results matches my prompt. I don’t know if the problem lies in the clarity of my prompt or the image model itself. So, I asked Niji again with a slightly modified prompt.

Prompt: a ship with “ありがとう” written — ar 16:9 — s 250 — niji 6

This time, it successfully rendered the text “ありがとう” on the side of the ship, which was what I had expected from the first prompt. This inconsistency is a bit frustrating and highlights that while the text rendering has improved, it’s not yet perfect.

Here’s another example:

Prompt: a cute 3D character that is holding a card that says “こんにちは” — ar 16:9 — s 250 — niji 6

Awesome. The text “こんにちは” (which means “hello”) was well-written, and the font matched the overall theme of the image perfectly.

Multiple Characters

One of the major upgrades to the new Niji model is its ability to generate images with multiple characters. In the previous versions, attempting to have two characters interact in a single frame would often fail.

To test this improved capability, I used the following prompt:

Prompt: two anime characters pose for a game poster, futuristic game concept, cool pose — ar 16:9

It looks super cool. Three out of four variations were successful. Here’s an upscaled version of the third image.

Niji V6 vs Niji V5

If you’re curious how the previous model compares to the new model, take a look at the results below using a similar prompt:

Prompt: girl with white hair, one-eye closed, holding a sign that says “愛してる” — ar 16:9

As you can see in the comparison above, the V5 model was not able to render the text properly unlike the V6 model.

Final Thoughts

Overall, I am happy to finally see Midjourney release an update to its AI models after months of silence. While it’s not a major change, it’s still a welcome upgrade, especially the text rendering improvement. There’s still room for improvement though, as some of my experiments show the text rendering doesn’t always work properly.

Niji is still a fun tool, particularly for those who work on anime illustrations or graphic designs. The enhancements in image quality and detail, such as the improved structure of anime eyes, will likely be appreciated by artists in this space.

Anyway, I’m still eagerly waiting for the Midjourney team to finally unveil their AI video generator or AI 3D tool. When are they going to release it? Can’t wait to get my hands on those tools.

‍