Create Best Anime images with Anything V3 or V4

This article is about creating high-quality anime-style images using the “Anything V3” artificial intelligence (AI) model. The “Anything V3” model is a latent diffusion model that is specifically designed to generate anime-style images based on input prompts and danbooru tags.

The article likely covers a range of topics related to using the “Anything V3” model, including how to provide input prompts and select danbooru tags to achieve specific aesthetics and themes.

It may also provide tips and techniques for using the model to create professional-quality anime-style images. Overall, the article is geared towards individuals who are interested in using AI to create their own anime-style images, whether for personal enjoyment.

Styles

I will be explaining how to create these styles, where you want to generate landscape or characters, i will be giving some examples prompt and what settings you should use to generate certain aesthetic and also what negative prompt to us.

Traditional anime style:

This style is characterized by exaggerated facial features, such as large eyes and small noses and mouths. The characters often have slender bodies and exaggerated expressions. Landscapes in this style tend to be highly detailed and may include elements such as futuristic cityscapes or surreal landscapes.

This can produce some artifacts, but you can refine it by using a negative prompt. Sometimes it might produce some blurry results, but experiment with steps and try to lower it a bit.

In terms of negative prompts, usually a single one, such as bad anatomy, will fix major issues, but you can try others, such as

Negative prompt:

You can use this in any of the anime style.

(((deformed))), blurry, bad anatomy, disfigured, poorly drawn face, mutation, mutated, (extra_limb), (ugly), (poorly drawn hands), fused fingers, messy drawing, broken legs, (mutated hands and fingers:1.5), (long body :1.3), (mutation, poorly drawn :1.2), black-white, bad anatomy, liquid body, liquid tongue, disfigured, malformed, mutated, anatomical nonsense.

Realistic anime style:

This style is characterized by more realistic facial features and proportions, as well as a greater attention to detail in clothing and surroundings. Characters in this style often look more lifelike and may have more subtle expressions. Landscapes in this style may be more grounded in reality and may include elements such as detailed cityscapes or natural environments.

What I mean by “realistic anime style” is that it usually means an anime style that is not too exaggerated and has complex lighting and reflection, something like “Garden of Words” and “your name.”

Also, you have to understand that sometimes 20 steps work better than 50 or 80, and sometimes I find 10 or fewer steps work better, so it’s up to you; you always have to experiment with it.

Prompt:

Add this to end of your prompt:

(realistic reflection:0.8), (realistic lighting:0.4), (rim lighting:1.5) 

Chibi style:

This style is characterized by exaggerated, cartoonish proportions, with small bodies and large heads. Characters in this style often have cartoonish appearances and may have exaggerated facial expressions. Landscapes in this style may be more simplified and may include elements such as colorful backgrounds or exaggerated architecture.

This style has been used in lots of slice-of-life anime. Generating in this style can be tricky sometimes, but what I found is that adding this prompt to anything can produce a consistent result.

Prompt:

Add this to end of your prompt:

(Chibi style: 0.5) (Chibi face: 0.3) (innocent face: 0.3) 

Aesthetic anime style:

This style is characterized by a focus on aesthetics and visual appeal. Characters in this style may have more stylized or exaggerated features and may be depicted in a range of poses and expressions. Landscapes in this style may include elements such as pastel colors, dreamlike environments, and surreal elements.

Anything v3 has four major settings: steps, sampler, and CFG scale. All these are significant settings, and they may require some adjustment to create your desired style.

I’m not going to explain why these settings are good for certain styles, but rather which settings to use for which styles.

This setting produces lot of artifact, but sometime this can create good result, so experiment with some settings.

This style is hard to explain, but its mostly in concept art, I think it as Tokyo goal’s ending image.