CivArchive
    Good Hands LoRA - v0.1[Beta]
    Preview 7170112
    Preview 7170154
    Preview 7173407
    Preview 7170091
    Preview 7170273

    If using Adetailer , Controlnet or inpaint, we recommend using prompt good hands
    Using lora str is 0.6-0.8.
    Hand poses are not yet supported.
    Not recommended for use in realistic model

    Main tag : good hands,

    Image Train : 168

    Training Powershell info :

    • DIM/Alpha : 32/32

    • Train batch size : 4

    • Training step : 4310x4x1 = 17240 steps

    • Learning rate Unet : 1.5e-4

    • Text Encoder Ratio : 0.5

    • resolution : 512

    • mix/save_precision = BF16

    • optimizer_type = AdamW

    • GPU : RTX 4060 16GB 160W

    • Torch 2.2 + cu12.1 + Xformers

    • Highvram = False

    • Vram used : 9.8GB

    • Speed : 1.69it/s

    • Loss Train : 0.0834 [No Min SNR]

    • Loss Test : ???

    • Local minimum : ?? time

    • Global Peak minimum : ??

    • Global Peak Maximum : ??

    • UNet weight average strength: 0.011379

    • Text Encoder (1) weight average strength: 0.0093899

    Description

    FAQ

    Comments (4)

    eddieAnimehentaiMar 13, 2024· 1 reaction
    CivitAI

    A weight of 0.7 actually works perfectly! thanks you!

    698059Apr 20, 2024· 1 reaction
    CivitAI

    Works pretty well with my personal model (realistic) and "HEUN" sampler. (Imo, "HEUN" does generally a better job on hands and feet.) Thanks for your work, m8, *thumbs up. Hope we'll see future updates on it.

    LewdMagicianJul 19, 2024· 1 reaction
    CivitAI

    Sorry Sir. I want to sell images using this Lora

    TyomanatorJul 30, 2024· 1 reaction
    CivitAI

    I'va made a test of this LORA on different weight, and the only effect i noticed was that on higher weights character was more prone to show his hands to camera. Quality of hands wasn't affected.