This is Akimabayu from Puella Magi Madoka Magica. There are six types: magical girl, school uniform, knit cardigan, casual clothes, maid, and lab coat.

Since the training images are game illustrations, this time I used WAI V14 as the base model to resemble the anime. As a result, the design is quite unique. I consulted CHATGPT and optimized it, but this is the limit.

Warning!

To recreate the anime illustrations, be sure to follow the instructions below!

If you don't, the shading will look like the game illustrations.

If this seems long, just follow the sample and you'll be fine!

The positive and negative prompts presented below should be used together. Changing the content will prevent it from looking like an anime.

Positive prompt

((flat 2D anime, cel shading, minimal shading, TV anime style)), soft lighting, even lighting

If the prompt in (()) is omitted, it will become a game illustration. Never include soft lighting or even lighting in (()). The face will look a bit strange and will resemble the game illustration.

Negative prompts

Realistic, 3D, photo, messy shading, extra limbs, poor ???????, blurry, poor hands, malformed hands, extra fingers, missing fingers, fused fingers, harsh shadow, dramatic lighting, strong contrast, realistic lighting

When using a background, it may be a good idea to also add white background, simple background, blurry background, or bokeh. Using these on a white background can cause the shading to lose its animation quality, so do not use them on a white background.

To address the high probability of shadows on the body or face when generating the upper body

The positive prompts soft lighting and even lighting and the negative prompts harsh shadow, dramatic lighting, strong contrast, and realistic lighting prevent shadows from falling on the face or body when generating the upper body, or prevent silhouettes from appearing on a white background. Use them together.

Step count: 20 or less, CFG 4-5.5, Intensity 0.7 or higher

Increasing the number of steps increases the amount of writing and reduces the animation quality. Increasing the CFG can make the face and painting look unnatural. If you don't mind a lower level of fidelity, a level below 0.6 is acceptable.

Never omit the trigger word "mahou shoujo madoka magica"!

It's included in the caption to approximate the style of Puella Magi Madoka Magica, so omitting it will significantly reduce the accuracy of reproduction.

Since the training is done on a white background, the background may be weak, simple, or the face may look subtle. If the face looks subtle, add "detailed face" and "sharp focus" after the positive prompt. It should never be in parentheses, as it would make the character look less anime-like. It should not be in parentheses when the background is white, as it would make the face look a bit strange.

They look the most anime-like when dressed in gothic lolita or other clothing that is outside the scope of the study. Explaining why this is the case would require several times more text, so if you have any questions, just ask CHATGPT! You'll get a satisfactory answer!

It looks most anime-like when dressed in gothic lolita or other clothing outside the target of study. Explaining why this is the case would require several times more text, so if you have any questions, ask CHATGPT! You'll get a satisfactory answer!