CivArchive
    Preview 34774324
    Preview 34780144
    Preview 34774314
    Preview 34777631
    Preview 34781933
    Preview 34778719
    Preview 34780393
    Preview 34782303
    Preview 34774320
    Preview 34781167
    Preview 34777104
    Preview 34774313
    Preview 34784047

    Join The Tinkerer on Whop. Membership gets you early releases, private tools and a bunch of extra stuff.

    👉 Join on Whop
    💬 Join the community for support, free tools and early news on Discord


    This is the model you reach for when you want super clean, photo-realistic results.
    It’s sharp, powerful, and somehow just nails gorgeous images every single time.


    ⚙️ Recommended Settings

    Sampling method: DPM++ 2M SDE Karras 
    VAE: is already Baked In
    Sampling steps: 30+ Steps
    Resolution: 832x1216 / 896x1152
    CFG: 3-5
    Clip Skip: 1
    Upscale: 1.5 upscale (0.3 - 0.35 denoise)
    Upscalers: 4x_NickelbackFS_72000_G / 4x_foolhardy_Remacri / 4x_NMKD-Siax_200k

    ⚙️ Recommended Settings DMD2

    Sampler: LCM
    Schedule Type: Exponential
    Sampling Steps: 8/12 (12 for extra detail)
    CFG Scale: 1-1.5
    ⚡ These settings are not the same as the standard SDXL workflow — be sure to adjust your pipeline!

    💡 What’s the difference between FP16 and FP32 checkpoints?

    FP16 (half precision) uses 16-bit floating-point weights. It’s smaller in size, loads faster, and uses less VRAM - perfect for normal image generation. You won’t notice any visual difference for inference, so it’s ideal for everyday use.

    FP32 (full precision) uses 32-bit weights, which means it’s larger and heavier on resources. It doesn’t improve generation quality, but it’s better if you’re training, merging, or fine-tuning models, since it keeps the exact weight data without rounding loss.


    🧠 Prompting

    Example negative prompt:

    cartoon, illustration, anime, painting, CGI, 3D render, low quality, watermark, logo, label

    or use CyberRealistic Negative


    Optional Adetailer

    Adetailer model: face_yolov9c.pt
    If you only want the main face being refined set 'Mask only the top k largest' to 1.


    🔗 Links
    Backup location: HuggingFace


    Fuel the Project
    If this model gave you stellar results, consider buying me a coffee. Or two. Or, you know… a GPU.


    ⚠️ Heads-Up
    This model can make mature or sensitive content. You’re fully responsible for what you create, so use it responsibly (or if you can’t resist, at least be artistically irresponsible).


    💬 Want Better Prompts?
    Use this custom ChatGPT to auto-generate polished prompts for CyberRealistic XL:
    🔗 [Try it now on ChatGPT]

    Description

    There have been significant changes and improvements across multiple areas. I decided to skip versions 2.3 and 2.4 due to unsatisfactory results.

    FAQ

    Comments (10)

    1922354Oct 15, 2024· 2 reactions
    CivitAI

    Another excellent release by Cyberdelia!

    shaniakellyOct 16, 2024
    CivitAI

    perfect 10/10 for 3.0,

    i think i suck at prompting and im getting amazing results

    kefam35707994Oct 16, 2024· 2 reactions
    CivitAI

    Very good heading to become some fantastic model.

    Pros contra v1, v2 line: realism, prompt adherence.

    Needs progression in eyes (serious eye problem there is, even your main picture has eye problem), celebrity, lora character likeness( only 1/5 is good, the rest is not even come near to character).

    Hell, I eagerly waiting even for a 4.2 xl to be match your legendary sd 1.5 checkpoint.

    Keep it up, very good progress.

    gausssidorov928Oct 16, 2024
    CivitAI

    v3 -SDXL

    9/10 nice

    kvsh88Oct 17, 2024· 1 reaction
    CivitAI

    I think the 3.0 Version is lacking in the celebrity lora dept, im unable to get consistent likeness to real of celebrity loras as compared to 2.2

    ElectricDreamsOct 17, 2024· 6 reactions
    CivitAI

    Cyberrealistic is truly on the way to be the best SDXL checkpoint... but not yet.

    I've compared it with a bunch of the most famous alternatives and Cyberrealistic 3 wins hands down in the iq department. If you use it properly.

    What was that supposed to meant? image quality is far better with lower CFG as 1.5. The images with CFG 6 or 7... are not as realistic to say the least. Probably is way to bright for my taste.

    So I recommend using it with a low CFG.

    The cons: prompt adherence is better in other sdxl checkpoints, faces with loras a sometimes a little bit weird if its not a closeup (closeups are amazing though), lora likeness probably has a little bit of room for improvement.

    All in all, yeah, in the best top 5 sdxl checkpoints for sure. With time and some tweakings, it might be the best one in SDXL as Cyberrealistic is in the SD. 1.5 platform. Just not yet.

    I'll keep forward to the next update. I think it will blow some minds. Mine included.

    All in all, great job Cyberdelia.

    Cyberdelia
    Author
    Oct 17, 2024· 2 reactions

    I've noticed also the 'issue' with faces and CFG and am already working on a new version (3.1). Version 2.4 had fewer problems, so I’m considering either releasing that version as well or using it as the foundation for 3.1, with some additional training from version 3.0.

    ElectricDreamsOct 17, 2024· 1 reaction

    @Cyberdelia many thanks for the tip! there was no need, really! Yeah now I've noticed that the color palette at CFG 3 is bloody fantastic. The warm tones are spot on. It's very distinctive too because i haven't found any other checkpoint with such colour richness. Its on a league of its own. I think it's actually very close to perfect (i have a hard time saying something is perfect!). But yeah, the faces at medium distance is the big issue.

    Courth7Oct 17, 2024
    CivitAI

    Where can I find the "CyberRealistic_Negative_SDXL" embedding? Cant seem to find it :(