CivArchive
    Sulphur 2 Base Quants - Dev
    NSFW

    THIS IS THE OFFICIAL MODEL PAGE FOR FP8 QUANTS OF SULPHUR 2

    A sequel to sulphur 1 on ltx 2.3, more data, more training, etc etc. If you're wondering why this isn't on TenStrip's account, I was the original creator, this is the base model. TenStrip did however create another merge for this much more updated version.

    OFFICIAL BF16 PRECISION DEV MODEL IS HERE:

    Sulphur 2 base - v1.0 | LTX Video Checkpoint | Civitai

    MORE VERSIONS AVAILIBLE HERE (fp8, lora, etc):

    https://huggingface.co/SulphurAI/Sulphur-2-base/tree/main

    SUPPORT FURTHER VERSIONS:

    https://ko-fi.com/fusioncow


    DISCORD:

    https://discord.gg/GSXJhKZ9V

    Description

    This is the official fp8 quantized model by SulphurAI

    FAQ

    Comments (6)

    AnimaXxMay 17, 2026
    CivitAI

    Is this decent and quickish at text to video or is it more for image to video?

    S1LV3RC01N
    Author
    May 17, 2026· 1 reaction

    The base Sulphur 2 model is primarily T2V and since T2V does not have the input image that increases time taken to generate, I would say it is relatively fast. Just keep check on your VRAM usage and if you notice that second pass after upscale is taking veeery long time, consider lowering resolution or duration. valid resolutions below 1280x768/1280x704 are 1152x648 and 1024x576. For duration, the model is best around 10 second duration. If you check out some of the links you will find more info.

    delta45424155May 18, 2026
    CivitAI

    what's different between this and the fp8 on sulphur's main page?

    S1LV3RC01N
    Author
    May 18, 2026

    This upload was actually approved by SulphurAI officially and I was given permission to upload it which can be confirmed by links in the official model page description as well as the Huggingface model card.

    delta45424155May 18, 2026

    @S1LV3RC01N that's not what i asked... But cool. The listed size here is different from the model's page. So that's why I asked if there is a difference.

    BigSad11May 18, 2026
    CivitAI

    Getting this error when I try this model:
    shape '[1024, 1024, 3, 3, 3]' is invalid for input of size 26245464