CivArchive
    Belle Delphine [Hunyuan Video] - v1.0

    If you enjoy my work please leave a like or a review. Pics Vids of your work are also appreciated :)

    Description

    FAQ

    Comments (10)

    DocueiJan 5, 2025· 5 reactions
    CivitAI

    Quite Fitting as her face is somewhat AI generated effects.

    Hell, even I'm AI generated. I couldn't hel-

    snaps out of it

    phew, my soul almost got taken over by openai

    georgelukJan 9, 2025· 1 reaction
    CivitAI

    Great lora! Would you be able to include your config or dataset for the training? I'm having trouble training a good lora.

    JaneB
    Author
    Jan 9, 2025· 1 reaction

    I dont share most of datasets as I dont own any of the images

    superpooperJan 26, 2025

    @JaneB willing to share #of images, epochs, and what tool you used for training?

    JaneB
    Author
    Jan 26, 2025· 1 reaction

    @superpooper I use ~25 images for all the hunyuan loras and trained on diffusion-pipe. Steps: 4,375

    Epochs: 35

    superpooperJan 26, 2025

    @JaneB thank you so much, I use about the same images for 1200 steps, maybe this is why my Loras are unusable. I'm going to try matching closer to what you're doing

    JaneB
    Author
    Jan 26, 2025· 1 reaction

    @superpooper try increasing the learning rate a bit too

    superpooperJan 26, 2025

    @JaneB any recommendations? The learning rates are a bit tricky for me to understand

    JaneB
    Author
    Jan 26, 2025· 1 reaction

    @superpooper I think the default is 0.0002 (might be another zero there so verify that) I do 0.0005

    superpooperJan 27, 2025

    @JaneB I've tried .0005 LR with 1800 steps, I think I may have over fit it now, but it worked for me. Thank you for sharing (:

    LORA
    Hunyuan Video

    Details

    Downloads
    1,343
    Platform
    CivitAI
    Platform Status
    Deleted
    Created
    1/5/2025
    Updated
    5/6/2026
    Deleted
    5/23/2025
    Trigger Words:
    belle delphine

    Files

    belleHunyuan.safetensors

    Mirrors

    CivitAI (1 mirrors)
    TensorFiles (1 mirrors)