CivArchive
    Preview 27676716
    Preview 27678057
    Preview 27678156
    Preview 27678218
    Preview 27678293
    Preview 27678476
    Preview 27678601
    Preview 27679343
    Preview 27679406
    Preview 27679592
    Preview 27679798
    Preview 27679893
    Preview 27679957
    Preview 27679982
    Preview 27680007

    To Use it online visit https://tensor.art/models/789608249918891303/GC's-Real-Horny-Pro-V3

    PLEASE DO NOT EXTRACT LORA, MERGE OR MODIFY FROM THIS MODEL WITHOUT PERMISSION.

    ________________________________________________________________________________

    Latest update:

    Real Horny Pro V3 (Unet). You'll need vae and clips for it to work. Finetunes and enhancements. It's now more versatile and creative, NSFW or SFW.

    ________________________________________________________________________________

    Uploaded Real Horny Pro V2 (NF4). You'll need vae and clips for it to work.

    Real Horny Pro V2 (Unet). Significantly enhanced lora compatibility, better nudity and overall performance.

    • Please leave comments or feedbacks of any kind.

    • Please share your creations.

    • Please give me a like or even a buzz donation if you like this model...pretty please~~~~

    thank you~~~

    ________________________________________________________________________________

    Real Horny Pro Asian Cuties (Unet). This is derived from Real Horny Pro, with improved performance in realism and lora compatibility. But, as the title suggests, this version likes to draw asian girls and women. The reason being that I don't like the presentation of asian females in the original Flux Dev, and I trained a lora based on my aesthetics of asian feminity to merge into this model.

    ____________________________________________________________________

    Uploaded gguf version as requested by some people. Huge thanks to https://civarchive.com/user/pkmngotrnr

    for taking the trouble to convert and upload the gguf model.

    • Please delete the .safetensors after gguf afterdownload(gguf is not a permitted format of checkpoint to upload to civitai so this was a makeshift solution)

    • Please don't ask me any tech questions of the gguf model as I don't personally use them. I am only uploading as per requested.

    ____________________________________

    Uploaded NF4 version as requested by some people. Please note this is only unet so you need to load flux clips and vaes for it to work.

    This version is based on the Real Horny Pro version with a little fine-tuning. Please do expect some loss in precision and quality of image generation, as NF4s are pruned. But they do require less vram and run faster.

    _________________________

    The Pro version is best of the best!!

    I named this version "Pro" instead of "V3", because this merge is totally at another level. I think this is the real "flux-level" that I have been trying to achieve. take a look at my sample pictures and you'll know what I mean. these pics are NOT cherry-picked, they came from the first 20 txt2img generations using the model.

    For the last 24 hours I went back and forth, try and error, experimenting with different methodologies, combinations and formulas like a mad scientist. But when I finally came out with this merge I felt so much worth it.

    I won't go into details boasting about it, but it's really really good.Try it, and you will not be disappointed.

    Again, hats off to the real heroes that shared their excellent models and loras with the community, to make this merge possible.

    And my most sincere thanks to everyone that supported me and this model. Thank you.

    ______________________________

    V2 is here!!

    Major updates:

    1. The problem of overfitting is addressed. V2 is now a lot more stable, flexible and easier to use.

    2. Substantial improvement in img quality. This is also the benefit of fixing the overfitting problem, restoring the details, vividness and realism of the Flux model.

    3. Significantly increased compatibility of Flux loras, giving users more controllability over the use of single or stacked loras.

    4. Fine-tuned to perform even better in generating female nudity, with better anatomy too!

    5. As a trade-off, the presentation and ability to generate sex acts have been reduced. this is due to the reason that I could not find a perfect balance in-between. Also there are already a lot of good loras out there for all sorts of things. For now I will just keep experimenting.

    ___________________________

    First of All I'd like to thank all the creators on Civitai who made all these awesome models and loras, and sharing with the community, they did all the hardwork and deserve all the thanks and respect. By using some of their great models and loras(I would not go into details as there are many of them), after more than 70 trials in merging and tweaking, I present to you: Real Horny Flux.

    This is a fp8 model based on Flux D1 that includes all clips needed (all-in-one).

    Yes, as the name suggests, this is a really horny model.

    Good things:

    1. Easy-peasy for generating nudes, breasts, nipples and pussies, at a very consistent good output. can perform simple sex acts such as vagnial/missionary. Very good with female anatomy.

    2. Prompt flexibility. can handle simple prompts and complicated, LLM style long prompts equally well.

    3. Fairly good Lora compatibility, to achieve this was the most difficult part during the process. But after tons of fine-tuning, this model now likes loras as much as you do! You'll just need to adjust the strengths of each lora and see how it works best for you.

    4. User friendly. Just load it with your favourite app, comfy or forge, key in prompts and click generate.

    5. Versatility. Again, due to the good compatibility this model is able to perform many other tasks as flux sub-models should, hence you can use it for general purpose (but do expect nsfw contents that might randomly occur).

    Not so good things:

    1. apart from the simple sex acts mentioned above, it is quite challenging to produce other concepts such as paizuri or fellatio. But, because it is very lora friendly, you can always use the NSFW loras to help.

    2. When using with loras, accuracy in generating texts drop. But I understand this is the general problem with all Flux models. Just FYI.

    3. Occasionally you'll see nipples on clothes. If you really want to avoid it, try putting "fully clothed" or something like that in your prompts.

    Anyways, this is the first version of a pretty good NSFW mode (I think), that meets the main idea of being SIMPLE and EASY to use. Hope you guys like it! Thank you.

    PS. All Flux D license terms and conditions apply.

    Description

    Please delete the .safetensors after .gguf afterdownload(gguf is not a permitted format of checkpoint to upload to civitai so this was a makeshift solution)

    FAQ

    Comments (30)

    GUS_C3Sep 4, 2024
    CivitAI

    Your gguf Q8 model is corrupted

    Traceback (most recent call last):

    File "D:\AI\webui_forge_cu121_torch21\webui\modules_forge\patch_basic.py", line 38, in loader

    result = original_loader(*args, **kwargs)

    File "D:\AI\webui_forge_cu121_torch21\system\python\lib\site-packages\safetensors\torch.py", line 308, in load_file

    with safe_open(filename, framework="pt", device=device) as f:

    safetensors_rust.SafetensorError: Error while deserializing header: HeaderTooLarge

    During handling of the above exception, another exception occurred:

    Traceback (most recent call last):

    File "D:\AI\webui_forge_cu121_torch21\webui\backend\loader.py", line 270, in forge_loader

    state_dicts, estimated_config = split_state_dict(sd, additional_state_dicts=additional_state_dicts)

    File "D:\AI\webui_forge_cu121_torch21\webui\backend\loader.py", line 236, in split_state_dict

    sd = load_torch_file(sd)

    File "D:\AI\webui_forge_cu121_torch21\webui\backend\utils.py", line 26, in load_torch_file

    sd = safetensors.torch.load_file(ckpt, device=device.type)

    File "D:\AI\webui_forge_cu121_torch21\webui\modules_forge\patch_basic.py", line 54, in loader

    raise ValueError(exp)

    ValueError: Error while deserializing header: HeaderTooLarge

    File corrupted: D:\AI\webui_forge_cu121_torch21\webui\models\Stable-diffusion\realHornyPro_rhProUnetQ80Gguf.safetensors

    Forge has tried to move the corrupted file to D:\AI\webui_forge_cu121_torch21\webui\models\Stable-diffusion\realHornyPro_rhProUnetQ80Gguf.safetensors.corrupted

    You may try again now and Forge will download models again.

    During handling of the above exception, another exception occurred:

    Traceback (most recent call last):

    File "D:\AI\webui_forge_cu121_torch21\webui\modules_forge\main_thread.py", line 30, in work

    self.result = self.func(*self.args, **self.kwargs)

    File "D:\AI\webui_forge_cu121_torch21\webui\modules\txt2img.py", line 121, in txt2img_function

    processed = processing.process_images(p)

    File "D:\AI\webui_forge_cu121_torch21\webui\modules\processing.py", line 799, in process_images

    p.sd_model, just_reloaded = forge_model_reload()

    File "D:\AI\webui_forge_cu121_torch21\system\python\lib\site-packages\torch\utils\_contextlib.py", line 115, in decorate_context

    return func(*args, **kwargs)

    File "D:\AI\webui_forge_cu121_torch21\webui\modules\sd_models.py", line 494, in forge_model_reload

    sd_model = forge_loader(state_dict, additional_state_dicts=additional_state_dicts)

    File "D:\AI\webui_forge_cu121_torch21\system\python\lib\site-packages\torch\utils\_contextlib.py", line 115, in decorate_context

    return func(*args, **kwargs)

    File "D:\AI\webui_forge_cu121_torch21\webui\backend\loader.py", line 272, in forge_loader

    raise ValueError('Failed to recognize model type!')

    ValueError: Failed to recognize model type!

    Failed to recognize model type!

    reading metadata for D:\AI\webui_forge_cu121_torch21\webui\models\Stable-diffusion\realHornyPro_rhProUnetQ80Gguf.safetensors: AssertionError

    Traceback (most recent call last):

    File "D:\AI\webui_forge_cu121_torch21\webui\modules\sd_models.py", line 89, in init

    self.metadata = cache.cached_data_for_file('safetensors-metadata', "checkpoint/" + name, filename, read_metadata)

    File "D:\AI\webui_forge_cu121_torch21\webui\modules\cache.py", line 114, in cached_data_for_file

    value = func()

    File "D:\AI\webui_forge_cu121_torch21\webui\modules\sd_models.py", line 81, in read_metadata

    metadata = read_metadata_from_safetensors(filename)

    File "D:\AI\webui_forge_cu121_torch21\webui\modules\sd_models.py", line 259, in read_metadata_from_safetensors

    assert metadata_len > 2 and json_start in (b'{"', b"{'"), f"{filename} is not a safetensors file"

    AssertionError: D:\AI\webui_forge_cu121_torch21\webui\models\Stable-diffusion\realHornyPro_rhProUnetQ80Gguf.safetensors is not a safetensors file

    GC
    Author
    Sep 4, 2024· 1 reaction

    thank you for your feedback. please read the instructions. you need to delete the .safetensors part after .gguf. although I don't use gguf myself I understand it can't be loaded normally like a safetensor file. I think you need a special comfy node to use it. and Again, I only uploaded the model converted by someone else, as has been requested.

    GC
    Author
    Sep 4, 2024

    if you are using forge, download the nf4 or unet version.

    GC
    Author
    Sep 4, 2024
    CivitAI

    Hi all, it's highly recommended that you use the unet version of the Real Horny Pro, if you are not familiar with nf4 or gguf models. the Unet version runs well both in comfy and forge, and has the best results (IMO). and please read the instructions first before you decide which one to download and use. thank you.

    EfficArtSep 5, 2024
    CivitAI

    How to use this in Forge. I get Clip error. Please advice

    EfficArtSep 5, 2024

    I checked this link before asking your help

    still can you share me which exact models I need to download, as As per this info, I downloaded and added models to respective folders

    GC
    Author
    Sep 5, 2024

    @EfficArt I recommend downloading the Pro Unet version. and you just need to download the vae and the two clips as explained in the forge site and put them into the right folders. then in forge, make sure you check flux type, load the unet, load the vae and the 2 clips. then you should be good to go.

    jeremyginSep 5, 2024
    CivitAI

    Thanks for sharing this - great model. Has anyone had luck conserving likeness in character models with it? Seems it affects character models trained on fp16 dev a lot.

    GC
    Author
    Sep 5, 2024

    thank you. most character loras work fine for me. if you are using forge, make sure to tick the type of automatic(fp16 loras) in the Diffusion in Low Bits option.

    girlswithafrosSep 5, 2024
    CivitAI

    Need more work on the vaginas, Bro.

    GC
    Author
    Sep 5, 2024· 2 reactions

    WIP bro. wip.

    girlswithafrosSep 6, 2024

    @GC Weiners too LOL. Thx !

    ZomblexSep 22, 2024

    That's what she said

    xplay800i722Sep 5, 2024
    CivitAI

    could u link the right clip, t5 and vae for the nf4 version pls i´m getting weight errors

    xplay800i722Sep 6, 2024

    @GC hm i got these... could u maybe give me a simple workflow because something seems to be wrong with mine

    GC
    Author
    Sep 6, 2024

    @xplay800i722 I use forge, sorry. there's plenty of good workflows on civitai.

    xplay800i722Sep 7, 2024
    CivitAI

    does someone got this (nf4 version) working on confyui? because i found no way

    oubart_exbrat940Sep 11, 2024

    Seem it's incompatible with ComfyUI, reports it's unable to determine file type! Probably because it was converted on Forge?

    __Miura__Sep 7, 2024· 19 reactions
    CivitAI

    The girls of this checkpoint are all fat, even when prompted otherwise.

    Klimpen53Sep 10, 2024· 2 reactions

    Nice

    Noob_eeSep 17, 2024· 2 reactions

    it is about time, flux are all skeletons, which are good too.

    roberto_baggioSep 12, 2024
    CivitAI

    @GC , I'm sorry, I probably done something wrong. I'm trying to download "Real Horny Pro" (aka v3) in fp8. I thought it would be on the "Real Horny Pro (Unet)" section, and the file to download says : Pruned Model fp8 (11.08 GB) - But when I click on the download link, it's downloading "realHornyProUnetNF4_realHornyProUnet.safetensors" ... which is the nf4 Unet, so were do I get the fp8 Unet ?

    Edit: it's weird, I get the same file when I Select "real Horny Pro (Unet)" and "Real Horny Pro NF4 Unet"

    Edit 2: from "Real Horny Pro NF4 Unet" the file I get is "realHornyProUnetNF4_realHornyProNF4Unet.safetensors", so I suppose this naming issue is coming from my browser...

    If it's the case, I'm sorry to bother you with that. And thank you for this model. :)

    GC
    Author
    Sep 12, 2024

    the files have been automatically renamed by the system. i know its confusing, sorry. what you need to do is look at the second part of the file, aka "_realHornyProUnet.safetensors" or "_realHornyProNF4Unet.safetensors". they are also different in sizes. the unet is 11gb plus, the nf4unet is 6gb plus. the nf4 requires less resouces and runs faster, but a bit less in precision and performance. if you have a powerful setup with more than 12gb vram i'd recommend the unet(11gb).

    hope it helps.

    roberto_baggioSep 12, 2024

    @GC Yes, I realized that after downloading both files :)
    Thank you very much GC.

    GC
    Author
    Sep 12, 2024

    @roberto_baggio no problem. have fun~~

    UnsignedLongshanksSep 12, 2024· 1 reaction
    CivitAI

    FP16 would make the most of this technology but I appreciate your work regardless

    GC
    Author
    Sep 12, 2024

    thank you for your kind words ~

    EricRollei21Oct 11, 2024

    also interested in fp16 unet. Thank you!

    Checkpoint
    Flux.1 D
    by GC

    Details

    Downloads
    1,394
    Platform
    CivitAI
    Platform Status
    Available
    Created
    9/4/2024
    Updated
    5/24/2026
    Deleted
    -

    Available On (1 platform)

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