CivArchive
    LCM-LoRA Weights - Stable Diffusion Acceleration Module - LCM for SDXL (Latest)
    Preview 9026062
    Preview 9026063

    LCM-LoRA - Acceleration Module!


    Tested with ComfyUI, although I hear it's working with Auto1111 now!

    Step 1) Download LoRA

    Step 2) Add LoRA alongside any SDXL Model (or 1.5 model, if using the SD 1.5 version)

    Step 3) Set CFG to ~1.5 and Steps to 3

    Step 4) Generate images in ~<1 second (instantaneously on a 4090)

    Basic LCM Comfy workflow attached as "Training Images" in zip format.

    Read more about the concept here;
    https://github.com/luosiallen/latent-consistency-model/blob/main/LCM-LoRA%20Technical%20Report/LCM-LoRA-Technical-Report.pdf

    Original HF link - SDXL; https://huggingface.co/latent-consistency/lcm-lora-sdxl/tree/main

    Original HF link - SD 1.5; https://huggingface.co/latent-consistency/lcm-lora-sdv1-5

    FAQ

    Comments (22)

    DuckFaceApr 2, 2024· 12 reactions
    CivitAI

    not work for me, deformed image

    ardeniusApr 3, 2024· 2 reactions
    CivitAI

    nice update, thank you

    LeeAeronApr 3, 2024· 6 reactions
    CivitAI

    A few questions:

    1. it's for LCM models only?

    2. how I can control enabling this feature? e.g. how I can control weight of this feature?

    HD1Jun 18, 2024· 17 reactions

    Weight should be 1. You can use it on every model but it works only on specific schedulers (lcm scheduler works always). I can recomend 7 steps and cfg 2.8 for good enough quality. I know you asked it 2months ago but the info can be useful for someone else.

    AugmentedRealityCatApr 5, 2024· 7 reactions
    CivitAI

    What's new with the "latest version" of LCM-SDXL exactly ? There is nothing new on the Huggingface repository, so I'm wondering where this "latest version" comes from, and what's new about it exactly.

    shapeshifter83Apr 5, 2024· 2 reactions

    the version posted here originally was out of date. If you check the dates, you will see that the first SDXL upload here was older than the huggingface upload. Seems this civitai listing has, at long last, been updated.

    audioscavengerMay 6, 2024· 2 reactions
    CivitAI

    This LoRA works with SD15 but the gain is limited for large pictures. You get similar results down from 15s to 7s, and cfg/steps from 6.5/20 to 3/8

    SesaruMay 9, 2024· 3 reactions
    CivitAI

    Excelente opción!

    urielmanx7May 12, 2024· 45 reactions
    CivitAI

    Scam.

    wzwowzwMay 14, 2024· 10 reactions
    CivitAI

    can be use in a1111?

    nindaimeraikageMay 31, 2024· 1 reaction

    yes

    mogJun 1, 2024· 6 reactions
    CivitAI

    a work for civitai

    hakaanJul 20, 2024· 11 reactions
    CivitAI

    For me better results are around 8 steps

    KNDSectorZAug 15, 2024· 7 reactions
    CivitAI

    Can this be used in Draw Things?

    SergigonSep 7, 2024

    sii si

    littlefluffyballDec 11, 2024· 15 reactions
    CivitAI

    If anyone still interested, it does work with any normal SD model, CFG 1.5, Steps 4-16, Sampler: Euler, LCM. Will speed up x2..x3.

    The issue - low CFG, makes prompts less effective, negative prompts not working.
    Images with CFG 1.5 and CFG 6 - will be different.
    LCM variant will be more "generic" even if you put more weight to the prompt, it will not help, as CFG still too low.

    gbrewtonDec 14, 2024· 10 reactions
    CivitAI

    2.5 cfg and 6 steps is the lowest I could go and still get usable results. 9 512x512 images in 6s with 4070ti

    tobycortesDec 21, 2024· 10 reactions
    CivitAI

    SICK, i use CFG 2.6 and use 30 Steps 512x768 hiresfix to 853x1280, it gives me great results and it doesnt hog my pc.

    tobycortesDec 21, 2024· 9 reactions
    CivitAI

    SICK, SDXL version same thing, CFG 2.6, 853x1280, 30 steps, great stuff !!!! really fast and not clogging my pc

    insertyourbenisplzJan 22, 2025· 17 reactions
    CivitAI

    This projects makes it viable to generate small images on CPU, a miracle indeed!

    Ninjun7Mar 13, 2025· 11 reactions
    CivitAI

    For me, if you take control away from the prompt, it's useless.

    scrunguscraigApr 25, 2025· 1 reaction

    And how exactly does this take control away from the prompt?

    LORA
    SDXL 1.0

    Details

    Downloads
    35,230
    Platform
    CivitAI
    Platform Status
    Available
    Created
    4/2/2024
    Updated
    5/13/2026
    Deleted
    -

    Files

    LCM_LoRA_Weights_SDXL.safetensors

    Mirrors

    HuggingFace (57 mirrors)
    TensorFiles (1 mirrors)
    TensorArt (1 mirrors)

    Available On (1 platform)

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