CivArchive
    Silverstar XL - v6
    NSFW
    Preview 17603775
    Preview 17603797
    Preview 17603867
    Preview 17603869
    Preview 17603897
    Preview 17603901
    Preview 17603929
    Preview 17603933
    Preview 17603938
    Preview 17603986
    Preview 17604110
    Preview 17604115
    Preview 17604150

    All of my preview images are txt2img, no hires fix, no face restore, no loras, no embeddings. I wanted you to see how this model generates.

    Edit: All v6 Previews are made with Lightning... I am finding the results better and faster than not using lightning, however still does great without it... Also a few did use the add detail lora.

    Face Restore (Codeformer) generally works well with this model, if needed/desired.

    Loras generally work well.

    My usual settings are Steps: 30, Sampler: DPM++ 3M SDE Karras, CFG scale: 3-7

    Works great with long and short prompts, lower CFG recommended on long prompts.

    LIGHTNING WORKS GREAT WITH THIS MODEL!

    I use this lora: https://civarchive.com/models/324115/sdxl-lightning-8step
    With EULER A, CFG 2, 16-20 Steps (Yes it is an 8 step lightning, but the higher step count reduces artifacts. Going over 20 seems to induce more artifacts, generally 16-20 seems to be the sweet spot for my models). I will include some generations using lightning to the gallery below.

    Introducing Silverstar XL!

    This is an SDXL model that I merged for my own aesthetic goals, primarily around the creation of young adult men. I found many models to either make them too attractive/girly, or too burly/strong without getting crazy on the prompt or negatives.

    Along the way I wanted to make sure I had creative options, capable of both photorealistic and artistic styles, and many different themes and subjects.

    Originally intended for personal use, I have been asked to share this, so here we are.

    As for NSFW, this model is not intended for NSFW. It may be possible to create, but the result may be not great. May in the future create a version for that.

    Description

    Skipping v5, all attempts were worse than v4FE at faces and hands

    Baked VAE.

    Seems to have better textures and more photographic than v4, lower amount of body horrors. (but it is still SDXL). Works great with Loras I tested, including NSFW ones.

    Note: Some NSFW capable models were used along the line but this is not an NSFW model, with the right prodding you might get something out of it, but especially for men use one of the many good Loras available.

    FAQ

    Comments (5)

    mscorpion7285Jun 29, 2024· 1 reaction
    CivitAI

    Well all I see are males in your showcase,so I wonder if this is a more male-focus model.

    But then you say its not good at nsfw with men,so its a sfw-male-focused model?

    SilverstarAI
    Author
    Jun 30, 2024· 1 reaction

    In short, its a photorealistic general mixed with a tiny amount of custom training, that I choose to focus on the male outputs because so many models are already really good at 1girl prompts.


    I mostly generate SFW 20-something men so I just wanted a good model for that for my own use and like to share it with the community. I have not been able to merge or train a model yet that is good with SFW and NSFW men... It is probably my lack of skill or knowledge. But I like my model's output and v6 and upcoming v7 both play well with the various lora's for the gens, but mixing them in kinda ruins the SFW qualities

    mscorpion7285Jun 30, 2024· 1 reaction

    Thx for your answer! Yeah I got your point. Looking forward to your next Versions.

    hugo_dockman_aiDec 12, 2025
    CivitAI

    I've tried this model for realistic artistic picture of male, but it didn't give very detailed good results. I've tried with a good refiner but one will be better for the first stage.

    SilverstarAI
    Author
    Dec 12, 2025· 1 reaction

    Hello! By AI standards this model is now ancient... but glad you gave it a shot! I no longer have a working A1111 which is what I used at the time, but can be used in comfy but may require vastly different settings, as Comfy seems to inference a bit differently to A1111.

    Also I never used it with any kind of refiner, all image examples were 1-pass with just my model and the lightning lora.

    If using comfy at this time I would highly recommend checking out Z-Image, its results are far superior and can run on nearly anything SDXL can...

    Checkpoint
    SDXL 1.0

    Details

    Downloads
    673
    Platform
    CivitAI
    Platform Status
    Available
    Created
    6/28/2024
    Updated
    5/28/2026
    Deleted
    -

    Files

    silverstarXL_v6.safetensors

    Mirrors

    CivitAI (1 mirrors)

    Available On (1 platform)

    Same model published on other platforms. May have additional downloads or version variants.