CivArchive
    FurryArt_Hyper4 - artistic
    NSFW
    Preview 13566883
    Preview 13566916
    Preview 13567037
    Preview 13576384
    Preview 13578438
    Preview 13567030
    Preview 13567028
    Preview 13583617
    Preview 13963393
    Preview 15575494
    Preview 15576412
    Preview 15875160

    This is a forked project of the "FurryArt" -> These models are HIGHLY EXPERIMENTAL

    * Merge Lab series *

    Note:

    v1.0 was renamed to "realistic"

    Realistic for realistic/semi-realistic images

    Artistic for "artistic" and some semi-realistic images

    Updated documentation text (NEW)

    Documentation:

    This is a merge model meged in a Hyper-SD. It is capable of producing sharp and detailed images starting with 4 steps and cfg 1 - 1.5

    How to use it.

    Sampling method

    1. DPM++SDE (for all the compositions)

    2. DDIM (for => 6 steps)

    3. EULER (A) (starting in 4 steps, but it will be better in => 6 steps)

    4. DPM++2m (from 6 to =>10 steps)

    5. Heun++2 (specifically for comfyui, for all the compositions)

    6. CFG++ (works well with comfyui, but you will need to test it with auto 1111)

    Schedule type

    Uniform and karras (good results)

    SGM Uniform (more accuracy)

    Align Your Steps (AYS) for 10 steps

    Sampling steps

    2 will produce a smooth but well-formed image

    4 standart step, will produce a sharp image for almost any composition

    6 for complex backgrounds and some complex compositions

    8 - 10 for very complex compositions and upscales

    10+ to use weighted styles or to do as before

    Styles (optional- specificaly for the "realistic" model):

    To use styles you need to add weight and negative prompts like “3d, realistic”

    Final words:

    Many thanks Gungie for for making this project a reality, I've learned a lot from you!!! 🤗

    Description

    This is a artistically oriented merge model meged in a Hyper-SD. It is capable of producing sharp and detailed images starting with 2* or 4 steps and cfg 1 - 1.5

    * better with 4 steps

    FAQ

    Comments (3)

    DitchWolfMay 25, 2024· 2 reactions
    CivitAI

    Both of these are incredible, it even forms eyes and mouths better than standard SD 1.5. Is Hyper 1.5 just a direct improvement over 1.5?

    VeerGeerMay 25, 2024· 2 reactions

    The regular 1-step LoRA and 12-CFG LoRA can be used as a generic enhancer, so kind of, yes.

    A_B
    Author
    May 26, 2024· 2 reactions

    So, hypersd is a model that facilitates the calculations to generate images,

    in my case I just merged a "LoRa hypersd" on my checkpoints

    and now as a result, I get detailed images and also with few steps!

    "Is Hyper 1.5 just a direct improvement over 1.5?"

    Well, no, "Hypersd" is an amazing LoRa model

    Checkpoint
    SD 1.5 Hyper
    by A_B

    Details

    Downloads
    165
    Platform
    CivitAI
    Platform Status
    Available
    Created
    5/25/2024
    Updated
    4/30/2026
    Deleted
    -

    Files

    furryartHyper4_artistic.safetensors

    Mirrors