CivArchive
    Realistic Tornado (Wan & Hunyuan Video LoRA) - v1.0

    Join our Discord community where you can generate videos, share ideas and request LoRAs, to push the bounds of the technology!

    WAN +HUNYUAN v1.0 :

    Trained on 50 real clips of tornadoes (each captioned separately), most of them 5 seconds in duration.

    The trigger word is: t0r54d0 tornado

    Wan 14B T2V Recommended Settings:

    LoRA strength = 1.0

    Embedded guidance scale = 6.0

    Flow shift = 5.0

    Hunyuan Recommended Settings:

    LoRA strength = 1.0

    Embedded guidance scale = 7.0

    Flow shift = 17.0

    Check out all our LoRAs on our page: https://app.remade.ai/

    Prompt Recommendations:

    Best to keep the prompt relatively simple, here are some examples that have worked well for me:

    Example 1: 'A t0r54d0 tornado spins over an empty carnival. A single Ferris wheel slowly turns as the wind picks up.'

    Example 2: 'A massive t0r54d0 tornado made of shimmering golden sand spins wildly in the middle of an ancient desert. The storm’s core glows, revealing glimpses of ancient ruins and statues being unearthed and swallowed again.'

    Example 3: 'A swirling t0r54d0 tornado made of countless books and papers spins wildly, pages flying in all directions. Titles and covers flash by as ancient knowledge is sucked into the sky, only to disappear into the vortex.'

    Let me know if there are any questions, I'll be happy to help!

    Description

    Version 1.0

    FAQ

    Comments (6)

    3427221Feb 21, 2025· 1 reaction
    CivitAI

    your models are pretty good, but could you train them in rank 32 (or even rank 16) instead of 64 ? they would be half or a quarter of the size, and more or less as good, maybe just a touch less precise at most

    robertdFeb 22, 2025

    Thanks for pointing out why this lora is so much better (cleaner and cripser) than most of the other loras. I definitely prefer size of 64 to blurriness of small size loras I see on this site.

    3427221Feb 22, 2025

    @robertd rank 32 is not automatically blurry, you could barely see any differences between a 32 and 64 with the same training dataset, even some rank 16 are sharp, depend of the training dataset mostly, while the file size is clearly different, 600+ vs 300 for a rank 64 over rank 32, about 150 for rank 16

    zengrathFeb 21, 2025· 2 reactions
    CivitAI

    I appreciate the effort into making unique lora's like this. it's kind of niche so i don't think many people will use it, but it's a nice to have! Maybe some will do creative things with loras like this.

    Remade
    Author
    Feb 21, 2025· 1 reaction

    Yeah that's the hope, just wanted to try some really cool ones!

    thefoodmageFeb 21, 2025

    I will use it! Great concept :D

    LORA
    Hunyuan Video

    Details

    Downloads
    327
    Platform
    CivitAI
    Platform Status
    Available
    Created
    2/21/2025
    Updated
    5/15/2026
    Deleted
    -
    Trigger Words:
    t0r54d0 tornado