CivArchive
    [Krea2 | moss] Nanmokaken Lora (んも) - moss v0.5
    NSFW
    Preview 145988004
    Preview 145988007
    Preview 145988009
    Preview 145988011
    Preview 145988014

    Artist style for krea2 | moss

    trigger: @nanmokaken

    Recommended settings: CFG=1, euler simple (Krea 2 Turbo); CFG=3 (moss image)

    Recommended resolution: mostly 1000~2000 px

    Recommended model: Krea2 Official Turbo (int8/fp8/bf16 are tested) | moss image base (Trained on base, works also for SFT (higher intensity needed))

    Description

    Work for both moss image base and sft

    The moss model seems to have some problem in drawing hands and sometimes have body unstability. Higher resolution and definite description helps in some cases.

    FAQ

    Comments (2)

    cutetodeath78409597Oct 18, 2026
    CivitAI

    how does moss take to lora training?

    KaleidRuby
    Author
    Oct 18, 2026

    I trained it on MOSS-Image Base with ostris/ai-toolkit (it already supports MOSS): 60 images, rank 32, lr 1e-4, batch 2 across two 5090s. It’s light to train — ~1 s/step with the DiT in bf16 and ~16 GB per card — and the style was fully picked up by around 1,600–2,000 steps; later checkpoints looked basically the same.

    LORA
    MOSS Image

    Details

    Downloads
    306
    Platform
    CivitAI
    Platform Status
    Available
    Created
    10/18/2026
    Updated
    10/27/2026
    Deleted
    -
    Trigger Words:
    @nanmokaken

    Files

    nanmokaken_moss_base_v1.safetensors

    Mirrors