CivArchive
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined

    THANK YOU FOR 500 DOWNLOADS AND 50 LIKES!

    UPD: October 8th, 2024 - V2 is released. Description is fully updated! Consider reading it!

    A complex merge of different checkpoints and LoRAs with Dark Anime Style Pony in its base.

    Differences from v1 Pro: just everything is better. Backgrounds, details, style - you name it.

    To get results that are looking the same as you see on previews - follow that workflow:

    1. Download upscale models (4xReal_SSIR_DAT_GAN, 2xLexicaRDBNet_Sharp, 4x_NKMD-PatchySharp_240k, 4x_NKMD-UltraYandere_300k, 1x-ITF-SkinDiffDetail-Lite). Place them into folders according to model type. You can use any of those as your main Hires. fix model (except for 1x one, use it for post-upscale). Find them on OpenModelDB or download from here (GDrive). Also, using RealESRGAN 4x Anime 6B is also a decent option, both for main Hires. fix model and secondary upscaler.

    2. I highly recommend using Stable Diffusion WebUI Forge. It has a parameter located in "Compatibility" called "Try to reproduce the results from external software". Choose "Diffusers" or "DrawThings" there - those two lead to some of the best results.

    3. In "UI alternatives" tab enable "Hires fix: show hires checkpoint and sampler selection" and "Hires fix: show hires prompt and negative prompt".

    4. In "Postprocessing" tab add "Upscale" to "Enable postprocessing operations in txt2img and img2img tabs".

    5. No additional extensions are required, all those built-in to Forge are good to go.

    6. Restart sampler is your friend and ally. It's almost required to use Restart. Schedule type can be set to automatic for both first pass and hires fix. Or choose one that you like the most and see what will happen :D

    7. I also recommend generating in higher resolutions. 2048x1536 works really well with "close-up" in positive prompt. But 1344x1024, 1536x1152 - also work and this is a key and a secret to crisp, detailed image.

    8. Hires. fix is not just recommended, it's required, too. Or if you don't like too detailed images, disable hires. fix, who am I to tell you.

    9. High CFG scale also works pretty well and leads to better images most of the time, but you might get some shitty generations before you get the one you'll love.

    10. "score_9, score_8_up, score_7_up, smooth, absurdres" is enough to get really good and detailed picture, adding more quality tags might lead to shitty outcome. You have been warned.

    Enough text! Now just look at those screenshots and set everything exactly the same (or change upscale models if you want to) to get results that look similar to what you see on previews.

    You might need a REALLY good PC to generate with a quality like that, or else it'll take an eternity to end. 2048x1536 pic with all hires fixes and postprocessing takes up to 6 minutes on a single RTX 3090. How it will work on your PC - I have no idea. Will test on my laptop with RTX 3050 Ti and add info there later.

    If you have any questions about this model, please leave a comment below - I read everything!

    Also leave a like, please! Finetuned version of that model is going to be released soon.

    With <3 from DigitalGarbage

    p.s. yes I love hoyoverse characters and you can do nothing about it

    Description

    Better anatomy, better details compared to V1X and older models.

    FAQ

    Checkpoint
    SDXL 1.0

    Details

    Downloads
    236
    Platform
    CivitAI
    Platform Status
    Available
    Created
    9/17/2024
    Updated
    4/30/2026
    Deleted
    -

    Files