- Why Midjourney fights you on anime
- The prompt structure that works
- A worked example, start to finish
- Keywords worth knowing by name
- Parameters that change the output
- The uncomfortable bit nobody puts in these lists
- Common mistakes that waste your credits
- When Midjourney isn't the right tool at all
- Frequently asked questions
The short version: Midjourney wasn't built as an anime model, so getting real anime style art out of it means fighting its default painterly, semi-realistic bias with specific keywords, style references and parameter settings rather than just typing "anime style" and hoping. I'll give you the exact prompt structure, the parameters that matter, and a worked example so you can see the difference a few words make.
Why Midjourney fights you on anime
Midjourney's training data leans heavily toward concept art, photography and painterly illustration. That's great if you want a moody fantasy landscape. It's a problem if you want something that looks like it came out of a Kyoto Animation production or a Shonen Jump cover. Type "anime girl" on its own into Midjourney and you'll usually get something that's anime-adjacent: big eyes, stylised hair, but with realistic shading, odd proportions, or a painted texture that no actual anime uses. That's the gap nobody tells you about upfront. You're not just adding a style word, you're correcting the model's instincts.
This matters because most people's first three or four attempts end up in the bin, not because their idea was bad but because the prompt structure was vague. If you've already read my guide to using Midjourney from first prompt to finished image, you'll know the basic workflow. Anime just needs a tighter version of that discipline.
The prompt structure that works
Here's the order I'd build an anime prompt in, every time:
- Subject and action (keep it concrete, not abstract)
- Specific anime style descriptor (not just "anime")
- Art medium reference (cel shading, flat colour, line art)
- Lighting or mood word
- Style reference parameter (--sref)
- Aspect ratio and version parameters
A weak prompt looks like this: anime girl standing in a city, anime style. A strong one looks like this: teenage girl with short silver hair standing on a rooftop at dusk, wind blowing her school uniform, 90s anime style, cel shaded, flat colours, bold outlines, warm sunset lighting --ar 2:3 --niji 6.
That last part, --niji 6, is the single most useful thing in this whole post. Niji is Midjourney's dedicated anime model, trained specifically on anime and manga aesthetics, and it exists precisely because the main model struggles. If you're serious about anime output, stop fighting the default Midjourney model and switch to Niji from the start. People write entire articles about anime prompting without mentioning this, which is a bit like writing about French cooking and never mentioning butter.
A worked example, start to finish
Say you run a small webtoon studio and need a cover image for a slice-of-life romance series. Here's how I'd build it up over four attempts rather than expecting one perfect prompt.
- Attempt 1: two teenagers sharing an umbrella in the rain, anime style --niji 6. Result: pleasant but generic, could be any shoujo series ever made.
- Attempt 2: add medium and era detail: two teenagers sharing an umbrella in the rain, 2000s shoujo manga style, soft pastel colours, delicate line art, blushing expressions --niji 6 --ar 3:4. Result: much closer, but the background is flat and empty.
- Attempt 3: add environment and lighting: ...cherry blossoms falling, neon shop signs reflected in wet pavement, golden hour light breaking through clouds --niji 6 --ar 3:4 --stylize 250. Result: this is cover-worthy. The stylize bump pushes Midjourney toward more artistic, less literal interpretation, which suits anime better than the default 100.
- Attempt 4 (optional): lock in consistency with a style reference image of an existing illustration you like using --sref plus the image URL, so future panels match the same linework and colour palette.
That's three or four iterations, five to ten minutes of work, and you've gone from generic to usable. That's the realistic pace, not the "perfect on the first try" fantasy a lot of tutorials sell you.
Keywords worth knowing by name
Generic words like "anime" or "manga" do almost nothing on their own because they cover eighty years of wildly different visual styles. Be specific instead:
- Studio or era references: "90s anime style", "early 2000s shonen", "modern digital anime" each pull different colour and line treatments
- Shading terms: "cel shaded", "flat colour", "soft gradient shading"
- Linework terms: "bold black outlines", "thin delicate linework", "screen tone shading" (for manga black and white look)
- Composition terms: "dynamic action pose", "dutch angle", "close-up portrait shot"
If you want a broader library of these descriptors beyond anime, my guide to art styles for AI prompts has copy-paste phrases for dozens of other looks too, which is useful if a client wants anime for one asset and watercolour for another.
Parameters that change the output
Beyond --niji 6, three settings do most of the work:
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
- --stylize (0 to 1000): higher numbers push toward more artistic, less literal results. For anime, 150 to 400 usually hits the sweet spot. Below 100 and you lose the stylised feel; above 500 and faces start to distort.
- --ar (aspect ratio): 2:3 or 3:4 for character portraits, 16:9 for scene or environment shots, 1:1 only for icons or avatars.
- --chaos (0 to 100): keep this low, under 20, for anime work. High chaos gives you wild variation between the four grid images, which is the opposite of what you want when you're trying to nail a consistent character.
The uncomfortable bit nobody puts in these lists
Midjourney, even with Niji, is still bad at consistent characters across multiple images. If you need the same character in twenty panels for a comic, wearing the same outfit with the same face, Midjourney alone will fight you the whole way. Each generation is a fresh roll of the dice unless you use seed numbers and style references religiously, and even then you'll get drift. This is one of the genuine limits of the tool, not a prompting skill issue, and anyone selling you a "perfect character consistency" prompt formula is overselling it. For that specific job, other anime-focused generators handle character locking better, and it's worth knowing the difference before you commit a whole project to one tool. I've covered some of those options in my piece on using an AI generator to create anime style artwork, and separately looked at what NovelAI includes beyond its free trial if you're weighing up dedicated anime tools against Midjourney.
Common mistakes that waste your credits
A few patterns I'd flag if you're new to this:
- Stacking too many style words at once ("anime, manga, cel shaded, watercolour, oil painting") confuses the model rather than refining it. Pick one coherent style and commit.
- Forgetting --niji 6 and wondering why the result looks half-realistic. This happens constantly.
- Over-describing faces in text when a style reference image would do the job in one parameter.
- Not setting aspect ratio before generating, then cropping afterwards and losing composition balance.
If writing the prompt itself is the sticking point rather than the parameters, a tool like my free AI image prompt generator can help structure the subject, style and lighting elements before you even open Midjourney, which saves a fair few wasted credits while you're learning the ropes.
When Midjourney isn't the right tool at all
If your actual goal is marketing imagery rather than illustration real, anime style is a strong visual hook for certain audiences (gaming, youth brands, Japan-adjacent products) but it's worth checking whether a broader image generator suits the wider campaign better. I've written about where to find DALL-E style AI image generators for marketing if anime is just one piece of a bigger visual content plan rather than the whole brief.
Frequently asked questions
Do I need Niji mode for anime prompts in Midjourney?
Not strictly, but you should use it. The standard Midjourney model can produce anime-adjacent images, but Niji 6 was trained specifically on anime and manga aesthetics and gets you there in far fewer attempts.
What stylize setting works best for anime style art?
Somewhere between 150 and 400 for most anime prompts. Lower than that and the image loses its stylised edge; higher than 500 and faces and proportions start to warp.
Why do my Midjourney anime characters look different in every image?
Midjourney doesn't guarantee character consistency by default. Using the same seed number and a --sref style reference helps narrow the variation, but for long-form comic or webtoon work you may need a dedicated anime generator built for character locking.
Can Midjourney copy a specific anime art style exactly?
It can get close with named descriptors like "90s shonen style" or "modern shoujo digital art" plus a style reference image, but it won't replicate a specific studio's house style precisely since it wasn't trained on labelled studio data.
Run a business, a brand or a blog? I feature tools, products and expert guest posts for an audience of 15,000. See how it works