CivArchive
    Shigeru Sudō Style 須藤 しげる - MOSS
    NSFW
    Preview 147993062
    Preview 147993220
    Preview 147993228
    Preview 147993305
    Preview 147993333
    Preview 147993336
    Preview 147993344
    Preview 147993373
    Preview 147993405
    Preview 147993501
    Preview 147993621
    Preview 147994091
    Preview 147994116
    Preview 147995632
    Preview 147995662

    This model is trained on illustrations by Shigeru Sudō (†1946). Sudō's style is Lyrical Painting (叙情画 jojōga), with works featuring in Japanese girls' magazines spanning from around 1926 to 1940.

    This first version was trained using Flux Gym with downscaled 512p images, and while it captures the style well, I would like to try again with a higher resolution. The first version also retains white borders from the training images, which you may or may not like. I will probably remove the white borders from later versions. Lastly, since the training images all have people showing their faces, it maybe take some finesse to get a character to face away from the viewer.

    On FLUX Dev, increasing the LoRA strength to 1.2 or 1.3 may help. On FLUX Schnell, it may help to kick the LoRA strength up as high as 1.5 to really get the rough coloring and prevent FLUX Schnell from increasing the contrast too much.

    https://ja.wikipedia.org/wiki/%E9%A0%88%E8%97%A4%E3%81%97%E3%81%92%E3%82%8B

    Translated excerpt from Japanese Wikipedia about Lyrical Painting:

    The term jojōga (叙情画) was coined by Fukiya Kōji (蕗谷虹児). For Fukiya, jojōga referred to paintings that express hidden sorrow outwardly. It often refers to illustrations of beautiful young girls, infused with the spirit of Taishō Roman (大正ロマン, Taishō Romanticism), created by artists such as Takehisa Yumeji (竹久夢二), Takabatake Kashō (高畠華宵), Fukiya Kōji (蕗谷虹児), Katō Masao (加藤まさを), Sudō Shigeru (須藤しげる), Matsumoto Katsuji (松本かつぢ), Tanaka Ryō (田中良), and Nakahara Jun'ichi (中原淳一), who illustrated for girls' magazines like Shōjo Gahō (『少女画報』), Reijokai (『令女界』), and Shōjo Club (『少女倶楽部』).

    https://ja.wikipedia.org/wiki/%E5%8F%99%E6%83%85#%E5%8F%99%E6%83%85%E7%94%BB

    Description

    Trained on the same dataset as the Krea 2 version, but with MOSS Image and 2 months of improvements in Fizgig, it seemed to come out better. Includes the trigger word "shigsustyle", but it seems unnecessary. It can help to specify "soft-shaded illustration printed on card stock". Strength 1 is fine, but strengths up to 1.4 help bring out more of the original art style.

    FAQ

    LORA
    MOSS Image
    by Bex

    Details

    Downloads
    86
    Platform
    CivitAI
    Platform Status
    Available
    Created
    11/5/2026
    Updated
    11/10/2026
    Deleted
    -
    Trigger Words:
    shigsustyle

    Files

    Moss Bex shigsustyle_v1-000037.safetensors