[PONY, Animagine] Inside of SHIBUYA SKY GALLERY & A View of Tokyo from There

update pony ver and add animagine ver

Addition and review of study images.

Lower the weights if necessary as they are over-learning.

English follows Japanese,

SHIBUYA SKYの屋内展望回廊「SKY GALLERY」内っぽい画像（＋そこから見える東京の景色っぽいもの）が作成できるLoRA

10個程作成して、フロアまである大きな窓の描写がもっとも良いと思われるものをアップロード

いくつかサンプル上げてるのでそっちも見てね

prompt

indoor lookout, black floor, window, cityscape from sky gallery

negative prompt (if U need)

chair, corner, couch, crowd, handrail, lantern, pillar, railing, sash, wooden floor,

既知の問題

(1) 教師画像が昼のものに偏っているため，一部のモデルで夜の画像がでにくい

*本LoRAとモデルの双方で"夜"の画像の学習が足りない為と思われる

(2) イラスト系モデルの場合，窓の低い位置でサッシや壁が生じる→handrail, railing, sash, wall 辺りをnegativeに加えることで解決する事がある

*SDのイラストモデルの場合、元々サッシが多く狭い窓が強く学習されていることの弊害

正確には窓枠を嵌める部分は框（かまち）というらしい

This LoRA can generate images that look like inside of the “SKY GALLERY,” the indoor observation corridor at SHIBUYA SKY (and a view of Tokyo that can be seen from there).

Create about 10 LoRA and upload the one that best describes the large window up to the floor.

I've uploaded some sample images, so take a look at those too.

prompt

indoor lookout, black floor, window, cityscape from sky gallery

negative prompt (if U need)

couch, chair, handrail, lantern, pillar, railing, sash,

Known issues

(1) The teacher images are biased toward daytime images, so it is difficult to obtain nighttime images for some models.

*This may be due to the lack of learning of “night” images in both the LoRA and the model.

(2) In the case of illustration-based models, sashes and walls occur at the low position of the window → sometimes adding handrail, railing, sash, or wall in the negative can solve the problem.

*The adverse effect of the strongly learned narrow windows with many sashes originally in the SD illustration model.