CivArchive
    Once upon a time (cartoon) style [FLUX] - v1.0
    NSFW
    Preview 57429583
    Preview 57429605
    Preview 57433777
    Preview 57429622
    Preview 57429943
    Preview 57429983
    Preview 57433356
    Preview 57433464
    Preview 57432110
    Preview 57432555
    Preview 57434280
    Preview 57434474

    Once Upon a Time... ("Il était une fois...") is a French educational animation franchise.

    Use ouat cartoon in your prompts. Use the V3 version

    Fine tuned with Astria AI

    A LoRA scale weight of 0.7-1.0 seems to be well (Click "Show more" for full image)

    I've made the Florence-2 captioned Dataset availble on Huggingface

    Description

    Fine tuned with Astria

    Use ouat cartoon in your prompts

    FAQ

    Comments (8)

    norod78
    Author
    Feb 13, 2025· 1 reaction
    CivitAI

    I think I should retry training for more steps with perhaps generated captions. Perhaps it will adhere better to the prompt

    PrimeDMFeb 13, 2025· 1 reaction

    and lower size 2/16 or 4/32 (rank 2-4 alpha 16-32)

    norod78
    Author
    Feb 13, 2025· 1 reaction

    @PrimeDM Good idea

    norod78
    Author
    Feb 16, 2025· 1 reaction

    Trained for 4500 steps with Automatic-Blip-Captioned text on a higher LR of 5e-4, reduced LoRA rank to 8. It still sucks.
    Might need to retry with manual captions

    PrimeDMFeb 17, 2025· 1 reaction

    @norod78 I use LR 0.0003 - 0.0001 almost always, on adafactor, adam and 4-8/32 Rank. And yes, the result depends greatly on the descriptions of the images.

    norod78
    Author
    Feb 17, 2025· 1 reaction

    @PrimeDM The V1 was 1e-4 which is 0.0001, then I tried 0.0005. Also, I had LoRAsa behaving well with rank 4 and 8. Somehow this specific dataset was not playing nicely. Anyway, re-did it with a huge rank and Florence2 captions and it finnaly looks good. So V3 incoming. If you wish to give it a go, I've uploaded the dataset and Florence2 captions here: https://huggingface.co/datasets/Norod78/OnceUponATime-florence2-captions

    But then again, the reader why this v3 iteration looks well is the Florence-2 based captions

    PrimeDMFeb 17, 2025· 1 reaction

    @norod78 Thanks, it turned out well, only the file size is very large. For flux it is clearly not necessary, perhaps only for some specific topics. I did not notice much difference when I changed rank/alpha and the size was only larger.

    norod78
    Author
    Feb 18, 2025· 1 reaction

    @PrimeDM I agree with you. Even in SDXL I managed to get stuff working even with rank size of 4. (Note that it is the rank size sets the size of the LoRA, the alpha is just a number which gets multiplied by the matrices, here is an awesome guide https://lightning.ai/lightning-ai/studios/code-lora-from-scratch). I wonder if all the creatures from the "Once-Upon-A-Time Life" in combo with the other versions made this spesific LoRA more challanging.

    LORA
    Flux.1 D

    Details

    Downloads
    108
    Platform
    CivitAI
    Platform Status
    Available
    Created
    2/12/2025
    Updated
    6/18/2026
    Deleted
    -
    Trigger Words:
    ouat cartoon

    Files

    ouat_cartoon-Flux-LoRA-2126374.safetensors

    Available On (1 platform)

    Same model published on other platforms. May have additional downloads or version variants.