CivArchive
    Manga style (Berserk) - Qwen v1
    NSFW
    Preview 108950022
    Preview 108950018
    Preview 108950011
    Preview 108950008
    Preview 108950004
    Preview 108950010
    Preview 108950013
    Preview 108950007
    Preview 108950015
    Preview 108950020
    Preview 108950009
    Preview 108950021
    Preview 108950017
    Preview 108950016
    Preview 108950019
    Preview 108950012
    Preview 108950006

    Klein 9B

    Showcase images :

    • Distilled model

    • 768x1152

    • Euler / normal / 4 steps / CFG 1

    • Lora strength of 1

    ZIB

    v1 :

    Still experimenting with Z-image Base training. Showcase images :

    • Euler / normal / 20 steps / CFG 5

    • Hires steps : 20 / NMKD_SIAX @0.2-0.5 / x1.5-x2

    • 768x1152

    • Lora strength 0.7 -1

    Qwen

    Qwen2512 v1:

    First lora trained on the new Qwen2512 (it still working with previous Qwen-Image and Qwen-Edit).
    For inference, I still prefer the first Qwen-Image model (BF16) (all the showcase images are done with this one).

    Seems pretty good but few things appears sometimes unprompted in my tests: artefacts on skin (like scars) and japanese onomatopoeia


    Qwen v1:

    I've trained a lot of different versions for this one : with big and small dataset, trigger words or not, high and low LR... I've still no idea what's good or not. 🙂

    Flux

    Had good results starting prompt with :
    'black and white Berserk manga style image of'

    and 0.8 strength most of the time

    v2 edit : Better dataset for better prompt adherence and flexibility

    Description

    FAQ

    Comments (6)

    JellaiNov 4, 2025
    CivitAI

    You said you trained a lot of different versions using a lot of different approaches. Which one did you land on? How did you approach it for Qwen?

    Yofaraway
    Author
    Nov 5, 2025· 1 reaction

    So I was not methodical enough (because I didn't plan to train it in so different ways) to compare every results and be sure it's the best one, but this one had 78 images in the dataset, and I used a LR of 0.0006 (20 repeats / 4 epochs / Cosine scheduler)

    JellaiNov 5, 2025

    @Yofaraway That learning rate is crazy high!

    Yofaraway
    Author
    Nov 5, 2025· 1 reaction
    JellaiNov 5, 2025

    @Yofaraway Their loras do look like some solid results... okay. I'm trying this next. Thank you!

    Rating_AgentNov 17, 2025· 3 reactions
    CivitAI

    i know the pain of training berserk lora. almost all of the dataset images has Guts in it, and kz of it every man looks like Guts and every woman looks like Casca, but your qwen lora is good! btw if you planning on training the v2 i can send you some not well known pics of berserk cards and illustrations from ps2 game

    LORA
    Qwen

    Details

    Downloads
    328
    Platform
    CivitAI
    Platform Status
    Available
    Created
    11/4/2025
    Updated
    5/20/2026
    Deleted
    -
    Trigger Words:
    Berserk manga style

    Files