Back to Discussions
Tips & Tricks

Experiment on light depiction in anime-style illustrations using Image V2

ねねむ
Translated to English·

On this site, among people who create anime-style illustrations, the number of those using Image V2 (hereafter, V2) has decreased considerably compared to before (recently, it's just me and about 3 others!).

Also, there is prior research on this site regarding the characteristics of other major models (such as the albums of "やまびこ | yamabico"). However, there is no description of V2...

So, in order to rediscover even one strength of V2, share it, and serve as a reference so that everyone can create higher-quality anime-style illustrations with V2, I am writing this down! 🖊️

〈Experiment Details〉

  • With the prompt and seed value kept the same, different expressions related to light were added to the prompt, and V2 was made to generate images.
  • The common prompt is "anime style, cute girl, long black hair, white blouse, black skirt, forest, flowering plants, and blue sky," and this image is the default.
    image

①Flash
image
The sun not only became an accent of the image, but the girl's hair also gained a gloss, making it more three-dimensional ☀️

②Sparkling
image
Particles of light were added among the trees, creating an image with gloss and gradation that is soft without being too dazzling ✨️

③Shining colorfully
image
The constituent colors of the background, the girl's hair, and the skirt increased, making the image more vivid 🎨

④Moonlight
image
The girl's hair has a subtle gloss, the background takes on more blue tones, and a sense of depth can be felt 🎑

From the above results, it was verified that, when generating anime-style illustrations, by using prompts that describe light expressions, V2:

  • Is capable of fine color differentiation
  • Is skilled at rendering three-dimensionality and vivid colors through that

Going forward, it would be nice to dig deeper into color expressions that V2 has difficulty generating, and into V2's distinctive qualities regarding light depiction when compared with other models.

If you all get tired of other models, please do make sparkling anime-style illustrations with V2 ~ (With so few fellow users, it leaves something to be desired 😂)

6 replies

こ
こばと🕊️_ダヴ(DOVE)·

Interesting research. I also like V2 and use it often. It beautifully reproduces light and color expressions. The biggest drawback is that it's prone to generation errors. Sometimes there would be three fingers, weird body proportions, and in the worst case, three legs 😂 But as you're testing, the soft color tones and bokeh are very charming 😂🕊️🌷

❤️1
ね
ねむOP·

Thank you for your kind comment!
You're right about V2's strengths and weaknesses. I also get troubled by generation glitches sometimes, but three legs is unusual 🤣
Still, I love V2's color expression, so I'll keep using it 👍

❤️1
ま
まれん2·

That's wonderful. I like things that shine colorfully✨
Keep up this momentum and promote V2 a lot🫶

❤️1
ね
ねむOP·

Thank you for your feedback!
I'd be happy if you could add this as a reference for the anthropomorphized profiles of the models currently being made😄
Also, I just created a masterpiece with V2, so please take a look at this too✨
image

❤️2
ま
まれん2·
@ねむ

I was just making sword images with V2 too, and it can create some lovely swords ⚔️ However, it's difficult to specify how they're gripped or handled...

👍1
ね
ねむOP·
@まれん2

I tried it too!
That's really true — it generates the sword itself beautifully, but instructions on how to hold it don't get through well….
Writing just "sword" actually seems to work better somehow 😅

😄1

Leave a reply

Join the community to share your thoughts and ideas.

Sign in to join the discussion