CivArchive
    XL-TCD-Imitation Of Reality (IOR) SAT - XL-TCD-IOR-SAT
    NSFW
    Preview 130190884
    Preview 130190895
    Preview 130190902
    Preview 130190923
    Preview 130190940
    Preview 130190946
    Preview 130190958
    Preview 130190961
    Preview 130191002
    Preview 130191004
    Preview 130191008

    Note: After a lot of testing, I think you can try the AYS schedulers, Gits, 11 and 32

    I want to start by saying that this model is quite experimental, remembering that it is an acceleration method that will work in 8 steps so please don't be too harsh in your judgment. Even so, I think it has given me very surprising results in many things.

    To begin with, what is this? It's an SDXL model based on the TCD acceleration system, a system that, unlike Turbo, Lightning, or DMD, delivers results much closer to a true base SDXL. If you want to know more about this, I've included two very informative articles.

    https://civarchive.com/articles/29730/sdxl-acceleration-methods-dmd-vs-tcd-vs-lightning-vs-turbo

    https://civarchive.com/articles/29775/tcd-for-sdxl-the-acceleration-method-that-feels-surprisingly-human

    The real advantage of using TCD is that, unlike DMD, for example, which only works with the LCM Sampler (which many complain about for its plasticky textures), here you can use a wide variety of Sampler-Schedule combinations.

    This model is experimental, as I said, since the TCD method is designed to run in Comfy with a specific sampler (TCD), but here I achieved a hybrid version so that it can be used not only in Comfy but also in Forge (as well as other methods, but for now I will only refer to those two).

    The parameters to be used are very important, so I recommend the following:

    • The VAE is BAKED

    • Upscaler: Im using 4x-ClearRealityV1 upscale by 1.5 Steps 10 denois 0.3 Hires CFG 1, i changed only the Hires Sampling to Euler A but the other parameter stay the same,

    • Width and height: 896*1152

    • CFG Scale: 1

    • Steps: 8

    • The sampling methods I recommend, although you can experiment with several, are: , DPM++SDE-Karras, DPM++SDE-SGM Uniform, DPM++SDE-Beta, Euler A-Beta, Euler Beta, DPM++SDE-AYS Gits

    The model works very well with CFG at 1, so there's no need to use negatives.

    What do I mean by saying that different samplers and schedules can be used? Judge for yourself.

    The prompt used for that image was: Vintage photograph of a Chinese woman in triad clothing in front of an arcade in Yokohama

    I really hope you can try it, give me your feedback, and help me improve. Keep in mind that it's one of the few SDXL models with a TCD system. Thanks.

    Description

    FAQ

    Checkpoint
    SDXL 1.0

    Details

    Downloads
    303
    Platform
    CivitAI
    Platform Status
    Available
    Created
    5/9/2026
    Updated
    5/13/2026
    Deleted
    -

    Files

    xlTCDImitationOfReality_xlTCDIORSAT.safetensors

    Available On (1 platform)

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