CivArchive
    Preview 10896282
    Preview 10896328
    Preview 10896520
    Preview 10896537
    Preview 10896711
    Preview 10896718
    Preview 10896538
    Preview 10896539
    Preview 10896714
    Preview 10896713
    Preview 10896717
    Preview 10896715
    Preview 10896716

    Pour les francophones, vous pouvez retrouver un guide d'utilisation sur:

    https://www.stablediffusion.blog/creaprompt-lightning-sdxl

    A1111 users, try https://github.com/tritant/sd-webui-creaprompt

    ComfyUI users, try https://github.com/tritant/ComfyUI_CreaPrompt

    CreaPrompt_Ultimate6

    Better colors, better lighting, better contrast, have fun!

    CreaPrompt_Ultimate4

    More realisme, more versatility

    CreaPrompt_Ultimate3

    This version is a merge of SD3 clips with CreaPrompt_Ultimate2. I used the same technique as @Adel_AI for Pixel Alchemy. I think this is my best version ever

    CreaPrompt_Ultimate2

    More realisme

    CreaPrompt_Ultimate

    This is a merge between CreaPrompt_Hyper_CFG_v2.0 (0.7) and Pixel Alchemy (0.3)

    https://civarchive.com/models/469255?modelVersionId=536516

    Usage:

    Euler Uniform 14 steps CFG 3,5

    But there are many different settings possibilities

    CreaPrompt_HyperSDXL_CFG

    V2.0

    Usage:

    sampler: Euler a SGM Uniform/Uniform

    steps: 6-25

    CFG Scale: 1-7.5

    To use a higher CFG, you must use higher steps

    V1.1 More realism, more versatility

    Hight CFG scale version, because of this, the rendering possibilities are multiplied

    For me the best settings is DPM++SDE Karras steps: 6 CFG scale 3

    Try Euler a Uniform (not SGM) steps: 10 CFG scale: 3

    Samplers:

    DPM++SDE Karras

    Euler a Uniform

    DPM++2S a Uniform

    Other settings tested:

    steps: 3 CFG scale: 2

    steps: 4 CFG scale: 3

    steps: 5 CFG scale: 4

    steps: 8 CFG scale: 5,5

    For more cfg, you need more steps

    CreaPrompt_HyperSDXL

    v1.2 More realism in 3 steps, much less artifacts, NSFW

    v1.1 A little less saturation and artifacts

    For detailed face, use Adetailer with this settings

    For more realism use:

    Steps: 11, Sampler: DPM++ 2S a, Schedule type: Karras, CFG scale: 1-1.5

    For more speed use:

    Steps: 3, Sampler: DPM++ SDE, Schedule type: Karras, CFG Scale: 1-1.5

    CreaPrompt_Lightning

    V1.3 more vibrance, more contrast

    V1.2 more lora merged

    The version 1.1 has much less artifacts, more vibrant colors and much more versatility.

    Usage tips:

    Steps: 4-7

    CFG scale: 1-3

    Sampler: DPM++ SDE Karras, DPM++ SDE

    For faces I use ADetailer with the model face_yolov8n.pt and a Inpaint denoising strength of 0,49

    V1.0 This checkpoint is oriented towards the Midjourney style portrait

    • Using advice

    Steps: 5

    Sampler: DPM++ SDE Karras

    CFG scale: 1

    Using Adetailer with Denoising strength: 0.49

    For more realism you can use this Lora

    https://civarchive.com/models/384163

    All image prompts come from the CreaPrompt collection, an extension for A1111/Forge

    https://github.com/tritant/sd-webui-creaprompt

    Description

    This is not a Lightning version but a Hyper-SDXL version

    FAQ

    Comments (6)

    goldmanApr 28, 2024· 3 reactions
    CivitAI

    An absolute game changer for me, many models I use produce great results on my M1 MacBook Pro but this one does it in 30 seconds instead of 2-3 minutes with no loss of quality...

    Highly recommended..

    Thank you so much.

    jice
    Author
    Apr 28, 2024· 1 reaction
    It's cool that this can help you, have fun
    LetTheBassDropApr 28, 2024· 3 reactions
    CivitAI

    Wow, what an amazing model. I have bad luck with any "lightning, hyper, turbo" SDXL models in the past but this one seems to be reigned in to where it doesn't introduce too many weird anatomy issues (Juggernaut's latest is a mess) and problems with image quality. This one responds well to prompts, maintains nice image quality, and can do a wide aspect ratio very well. Not sure why we train using vertical aspect ratios considering most media content is an a horizontal wide aspect ratio. My TV, monitor, and so on aren't oriented like a phone is and I don't do image generation from my phone hehehehehe.

    jice
    Author
    Apr 28, 2024
    Thank you for your feedback, I'm glad you like it, have fun
    WhatDaFAIMay 3, 2024· 2 reactions
    CivitAI

    I never liked the results of Lightning or Turbo-Models. Hyper is great and your Merged Model is amazing - with 8 Steps its giving me great results. Great Job - Thank you!

    jice
    Author
    May 3, 2024
    Thanks for your feedback, have fun
    Checkpoint
    SDXL Hyper
    by jice

    Details

    Downloads
    920
    Platform
    CivitAI
    Platform Status
    Available
    Created
    4/27/2024
    Updated
    5/19/2026
    Deleted
    -