CivArchive
    Anime Art Diffusion XL - alpha2
    NSFW
    Preview 1743709
    Preview 1743741
    Preview 1743704
    Preview 1743705
    Preview 1743706
    Preview 1743708
    Preview 1743710
    Preview 1743711
    Preview 1743757
    Preview 1743725
    Preview 1743723
    Preview 1743726
    Preview 1743729
    Preview 1743728
    Preview 1743731
    Preview 1743860
    Preview 1766192
    Preview 1766350
    Preview 1766347

    Anime Art Diffusion XL

    Check the version description below (bottom right) for more info and add a ❤️ to receive future updates.
    Do you like what I do? Consider supporting me on Patreon 🅿️ to get exclusive tips and tutorials, or feel free to buy me a coffee

    Kind of deprecated now, get AAM XL


    Raising from the ashes of ArtDiffusionXL-alpha, this is the first anime oriented model I make for the XL architecture. It will serve as a good base for future anime character and styles loras or for better base models.

    Should work well around 8-10 cfg scale and I suggest you don't use the SDXL refiner, but instead do a i2i step on the upscaled image (like highres fix).

    It's trained on multiple famous artists from the anime sphere (so no stuff from Greg Rutkowsky, but stuff like Sam Yang and Wlop for example). The model is oriented at producing stylized results.
    I also suggest you use words like 8k and high resolution because otherwise it tends to go too much low detail. Look at the examples.

    Enjoy!

    Description

    Based on SDXL1.0 but with the better VAE

    FAQ

    Comments (26)

    EDGJul 27, 2023· 14 reactions
    CivitAI

    Thank you for your contribution to XL, Lykon!

    2501Jul 27, 2023
    CivitAI

    What was your comfyui workflow?

    CallMeMaybeJul 27, 2023

    the workflow is included in the image

    2501Jul 27, 2023

    @CallMeMaybe I have literally tried to put the breast one into ComfyUI and it doesn't detect anything.

    Lykon
    Author
    Jul 27, 2023

    @virtualfix6885 workflow data is stripped by the cdn, but civitai stores it in the workflow button. Go to the generation data section of the image and click on "workflow"

    CAzVmN758Jul 27, 2023

    @Lykon Oh nice, never knew you could just copy text and then Ctrl+V the workflow into ComfyUI directly

    DEgaurdianJul 27, 2023
    CivitAI

    Is there anything special that we need to do to use this? Or is it just download and use like a normal older style models?

    Lykon
    Author
    Jul 27, 2023

    this is just a base model based on the XL architecture that's good at doing some anime artist styles. It will be my repository for anime model based on xl.

    WerryJul 30, 2023

    @Lykon Amazing work! Please do Evangelion 1995 anime style Lora for SDXL1.0

    Lykon
    Author
    Jul 30, 2023

    @Werry I've done it for sd1.5, is it really needed for xl? Won't work well at the moment I feel.

    datSatoJul 27, 2023· 2 reactions
    CivitAI

    Looking good!
    I'm really hoping we also get SDXL models that are more "true anime" (cartoony, far removed from reality, cute) rather than realistic and semi-realistic styles.

    Lykon
    Author
    Jul 27, 2023

    this is still required as a first step, I'll do that too, don't worry

    LostAtSeaJul 27, 2023· 6 reactions

    I often find that "true anime" style tends to be very ascetically unpleasingly compared to the Blender 2D look

    kudon44Jul 28, 2023· 1 reaction

    Yeah, classic styles like in the 90s animes and early 2000s would be cool since anime art was so much more diverse compared to now or the most common models.

    KorewaaiJul 29, 2023

    @Lykon How many images do you have to do finetuning?

    LostAtSeaJul 27, 2023· 13 reactions
    CivitAI

    And so it begins

    KimnaniJul 28, 2023· 1 reaction
    CivitAI

    is there a diffusers model for this?

    dunlurJul 28, 2023
    CivitAI

    Would you reccomend Clip skip 1 or 2 with this?

    Lykon
    Author
    Jul 28, 2023· 1 reaction

    it's XL, go with 1. Clip skip 2 was because of NAI, we didn't have that on XL.

    dunlurJul 28, 2023· 1 reaction

    @Lykon cheers

    KorewaaiJul 29, 2023

    @Lykon But why NAI used clip skip 2?

    Lykon
    Author
    Jul 29, 2023

    @Korewaai no idea.

    ai_degenxJul 29, 2023

    @Korewaai It was due to a mistake on their part when training with the clip numbers or something like that.

    Vanadium74Jul 28, 2023
    CivitAI

    what tool did you use for training? I have used a google colab for training on 1.5 but wasn't sure about SDXL training.

    syz840627903Jul 29, 2023
    CivitAI

    That's about vae,Do I need to use the original version of SDxl1.0 or not?

    Lykon
    Author
    Jul 29, 2023

    this one has its own vae.