CivArchive
    Ultimate DeepThroat & General NSFW - LTX2.3 Video LoRa - K3NK - LTX2.3 - v2.0
    NSFW

    • ᴛʜɪꜱ ʟᴏʀᴀ ɪꜱ ᴘʀᴏᴠɪᴅᴇᴅ ᴀꜱ ɪꜱ ᴀɴᴅ ɪɴᴛᴇɴᴅᴇᴅ ꜰᴏʀ ʟᴀᴡꜰᴜʟ ᴀɴᴅ ʀᴇꜱᴘᴏɴꜱɪʙʟᴇ ᴜꜱᴇ.

    • ɪᴛ ᴀᴄᴛꜱ ᴀꜱ ᴀ ɢᴇɴᴇʀᴀʟ-ᴘᴜʀᴘᴏꜱᴇ ʀᴇꜰɪɴᴇᴍᴇɴᴛ ʟᴀʏᴇʀ, ɪɴꜰʟᴜᴇɴᴄɪɴɢ ʜᴏᴡ ᴛʜᴇ ʙᴀꜱᴇ ᴍᴏᴅᴇʟ ᴇxᴘʀᴇꜱꜱᴇꜱ ᴘʀᴏᴍᴘᴛꜱ ᴀɴᴅ ᴄᴏɴᴄᴇᴘᴛꜱ ʀᴀᴛʜᴇʀ ᴛʜᴀɴ ᴘᴇʀꜰᴏʀᴍɪɴɢ ᴅɪʀᴇᴄᴛ ᴛʀᴀɴꜱꜰᴏʀᴍᴀᴛɪᴏɴꜱ. ᴛʜᴇ ʀᴇꜱᴜʟᴛɪɴɢ ᴏᴜᴛᴘᴜᴛ ɪꜱ ᴀʟᴡᴀʏꜱ ᴅʀɪᴠᴇɴ ʙʏ ᴛʜᴇ ᴘʀᴏᴍᴘᴛ ᴀɴᴅ ᴛʜᴇ ᴄᴀᴘᴀʙɪʟɪᴛɪᴇꜱ ᴀʟʀᴇᴀᴅʏ ᴘʀᴇꜱᴇɴᴛ ɪɴ ᴛʜᴇ ᴜɴᴅᴇʀʟʏɪɴɢ ʙᴀꜱᴇ ᴍᴏᴅᴇʟ.

    • ᴛʜᴇ ʟᴏʀᴀ ᴅᴏᴇꜱ ɴᴏᴛ ʀᴇʟʏ ᴏɴ ᴘᴀɪʀᴇᴅ ɪɴᴘᴜᴛꜱ, ʙᴇꜰᴏʀᴇ/ᴀꜰᴛᴇʀ ᴇxᴀᴍᴘʟᴇꜱ, ᴏʀ ᴛʀᴀɴꜱꜰᴏʀᴍᴀᴛɪᴏɴ-ꜱᴘᴇᴄɪꜰɪᴄ ᴛʀᴀɪɴɪɴɢ. ɪᴛ ꜰᴏᴄᴜꜱᴇꜱ ᴏɴ ɪᴍᴘʀᴏᴠɪɴɢ ᴄᴏʜᴇʀᴇɴᴄᴇ, ᴅᴇᴛᴀɪʟ, ᴀɴᴅ ꜱᴛᴀʙɪʟɪᴛʏ ᴀᴄʀᴏꜱꜱ ɢᴇɴᴇʀᴀᴛɪᴏɴꜱ, ɪɴᴄʟᴜᴅɪɴɢ ᴍᴀᴛᴜʀᴇ-ᴛʜᴇᴍᴇᴅ ᴏᴜᴛᴘᴜᴛꜱ ᴀʟʀᴇᴀᴅʏ ꜱᴜᴘᴘᴏʀᴛᴇᴅ ʙʏ ᴛʜᴇ ʙᴀꜱᴇ ᴍᴏᴅᴇʟ.

    • ɴᴏ ꜱᴘᴇᴄɪᴀʟɪᴢᴇᴅ ᴅᴀᴛᴀꜱᴇᴛꜱ ᴛᴀʀɢᴇᴛɪɴɢ ɪᴅᴇɴᴛɪᴛʏ ᴇxᴘᴏꜱᴜʀᴇ ᴏʀ ᴄᴏɴᴛᴇɴᴛ ʀᴇᴄᴏɴꜱᴛʀᴜᴄᴛɪᴏɴ ᴡᴇʀᴇ ᴜꜱᴇᴅ ᴅᴜʀɪɴɢ ᴛʀᴀɪɴɪɴɢ. ᴀʟʟ ꜰᴀᴄᴇꜱ ᴀɴᴅ ᴘᴏᴛᴇɴᴛɪᴀʟʟʏ ɪᴅᴇɴᴛɪꜰɪᴀʙʟᴇ ꜰᴇᴀᴛᴜʀᴇꜱ ᴡᴇʀᴇ ᴍᴀꜱᴋᴇᴅ ᴏʀ ᴀɴᴏɴʏᴍɪᴢᴇᴅ ᴛʜʀᴏᴜɢʜᴏᴜᴛ ᴛʜᴇ ᴛʀᴀɪɴɪɴɢ ᴘʀᴏᴄᴇꜱꜱ.

    • ʙʏ ᴅᴏᴡɴʟᴏᴀᴅɪɴɢ ᴏʀ ᴜꜱɪɴɢ ᴛʜɪꜱ ʟᴏʀᴀ, ʏᴏᴜ ᴀᴄᴄᴇᴘᴛ ꜰᴜʟʟ ʀᴇꜱᴘᴏɴꜱɪʙɪʟɪᴛʏ ꜰᴏʀ ɪᴛꜱ ᴜꜱᴇ ᴀɴᴅ ᴀɢʀᴇᴇ ᴛᴏ ᴄᴏᴍᴘʟʏ ᴡɪᴛʜ ᴀʟʟ ᴀᴘᴘʟɪᴄᴀʙʟᴇ ʟᴀᴡꜱ ᴀɴᴅ ᴘʟᴀᴛꜰᴏʀᴍ ʀᴜʟᴇꜱ.


    Ultimate Deepthroat - LTX2.3 Video LoRa - K3NK

    • V4 is ready, after weeks of training and testing different combination of previous loras.. it has two brand new datasets, also an image one ive made with my cumonface lora for klein9b to reinforce the penis shape (sideways sadly pov is not that great, also i didnt really try other nsfw aspects since the general nsfw loras from the firsts versions are being "diluted" by the new datasets..)

    • V3 contains a bit of Eros10 lora at 0.3 str + my bbc lora and g4gging loras, those loras have no audio, so is best to merge with previous lora adding new features like better color for black penises and so on

    • V2 contains sulphur lora merged at rank 77 with audio_to_video mode 1.0 str, stacking this lora with sulphur at high str can produce noisy audio

    • I trained plenty of my datasets and the results were no good, i was not getting penis anatomy, but when combining with key loras results are good, so i ended up making a merged model with loras like Penile Praxis, Rogue Missionary and Furry Lora with couple of my loras, some i2v only, others t2v that i trained for 24k steps, it was a bit overtrained but was great for using alongside others.

    • ive found the perfect balance (most of the times) to produce a nice looking penis and avoid no-clipping of the penis in the woman's mouth

    • I've being training a lot , testing, so ive decided to just share this one, no early access, free, so i'll really appreciate any kind of support being buzz or ko-fi

    Deepthroat Example Prompt

    cinematic scene witha woman in frame, when a penis enters from the side, only his lower body is visible, side view of his hips, thighs and legs, with a erected penis with testicles appears and she starts engaging in a deepthroat blowjob with that penis. She produces the following sounds as she swallow the penis: "GWA, GWA, GWA, GWA!!"
    
    She swallows the entire penis, her nose smashes agains the man's hips. She moves her head back and fort swallowing the penis, realistic proportions, natural lighting, cinematic composition

    Sex Example Prompt

    a nude woman lying on her back, a man's penis is at the bottom, penetrating the woman's anus, the man is thrusting his penis back and forth , penetrating deeply the woman anus, they're having fast phased anal sex

    Have fun! 😜

    ☕KOFI https://docs.google.com/forms/d/e/1FAIpQLScltzaQXTF2VFWRWVaqPSzJXYyNBF9hUQOLI8aBCyLrpCM1_w/viewform?usp=dialog

    Description

    This version contains some more of my lora versions, further refined the penis shape with loras trained with prodigy optimizer in the musubi fork, they are not perfect tho.. but when combined propperly gives greats results, also i merged sulphur lora as audio_to_video, seems o improve overall actions and sounds, even better that when used in a normal lora loader.

    Be aware Wan2GP users can have issues with this lora file, dont ask me why..

    FAQ

    Comments (47)

    wsx805758618May 10, 2026· 2 reactions
    CivitAI
    It's good, but it also has drawbacks; Asians experience facial drift when they lower their heads.

    WhyNaNWhenNyanMay 10, 2026
    CivitAI

    yes

    simonishereMay 10, 2026

    yes x2

    Ragamuffin20May 10, 2026· 1 reaction
    CivitAI

    not really judging anyone just wondering why you all seem to like none POV views.

    boobkake22May 11, 2026· 4 reactions

    Some of us don't need to imagine a person is doing a thing to us to enjoy it. Plenty of pleasure to be derived from watching others.

    toyder339May 11, 2026
    CivitAI
    First version works better for me. 2 not worth it.

    K3NK
    Author
    May 11, 2026· 1 reaction

    i mean, you can see in the example videos, how the throatfucking is wild and good, also the sounds,, but ok.. xD

    kronos1959777May 11, 2026· 1 reaction
    CivitAI

    I would avoid using distilled 1.1 to do your samples. the voice breathing in distortion is pretty bad. 1.1 is the worst.

    kronos1959777May 11, 2026· 2 reactions
    CivitAI

    this lora does not work in wan2gp. is this sulphur? you need to put that in the description somewhere. :(

    K3NK
    Author
    May 11, 2026· 1 reaction

    Is on the version description

    animart920May 11, 2026· 1 reaction
    CivitAI

    please make pussy licking lora.. 🙂 please please ...Or if there is any ltx 2.3 Lora, please let me know.

    K3NK
    Author
    May 11, 2026

    I'm on it, my cuni Lora, is at 3.5k steps, I'll run it a bit more, since t2v seems not ready yet..

    animart920May 11, 2026

    @K3NK Thanks Sir for reply ... Will this work in i2v as well?

    K3NK
    Author
    May 11, 2026

    @animart920 yea, idk about that Wan2GP tho.. everyone seems to have issues with my ltx loras when using that ..

    iammrrobot420May 12, 2026· 2 reactions
    CivitAI

    so if i only use wan2gp, i cant use this lora? wasted buzz on my part.

    ElGordoAIMay 12, 2026

    why not? if it's just a lora... and if it's for sulphur you can use it, the merge or just the lora on any quant

    iammrrobot420May 12, 2026· 1 reaction

    @ElGordoAI  i use wan2gp and it doesnt work on an ltx2.3 model sadly

    dillion1920May 26, 2026

    Ah, f3ck, again? god damnit.

    Trillium9May 27, 2026

    Bummer. Doesn't work for me on Wan2GP with ltx2.3, either. Glad it wasn't just me doing something wrong.

    sfractMay 12, 2026
    CivitAI

    good lord he's done it...

    djmarksmile912May 14, 2026
    CivitAI

    Guys, can someone give me a working assembly of Comfy UI, I'm already tired of going through them and every now and then one of the nodes doesn't work.

    Sexybeast10May 21, 2026

    I really enjoy Lycos Stability Matrix, and using Comfyui through there, very user friendly and easy to update, ALSO I am so Happy K3NK has a red civtai now to post all these amazing Loras on <3

    K3NK
    Author
    May 21, 2026

    @Sexybeast10 ill rather have a easily purchaseable yellow buzz website than this red one.. but i get your point xD

    Sexybeast10Jul 11, 2026

    Are you using comfyui on the civtai site? And if so, can I ask how much buzz ish it cost to make a 20 sec video?

    K3NK
    Author
    Jul 11, 2026

    @Sexybeast10  i do all locally with a 4090

    krush106107326May 15, 2026
    CivitAI

    Will my RTX 5060 16GB graphics card handle this model and video?

    K0DEXMay 17, 2026

    @krush106107326 Yes

    ElGordoAIMay 26, 2026

    try in wan2gp

    aureliusMay 27, 2026

    Yes, it can.

    Ragamuffin20May 25, 2026· 6 reactions
    CivitAI

    just a heads up, the file name is really wierd. idk if civ changed it or its fresh from the training but ltx23-ultimatedt-NSFW-sulphured_audio_v2_k3nk. safetensor is a little bit much

    dillion1920Jun 16, 2026

    I disagree, more people should do this.

    wqn999May 26, 2026
    CivitAI

    Hello everyone, I would like to ask a question. Regarding the inference of the NSFW video training set, which large language model do you use? Qwen3.5/Qwen3.6/Gemm4?

    K3NK
    Author
    May 26, 2026

    this uses gemma yeah

    psspsspsspssspssMay 26, 2026
    CivitAI

    Can we get a version that doesn't have sulphur merged? sulphur is my base these days, I don't need it twice included.

    K3NK
    Author
    May 26, 2026

    ive tested using eros10 checkpoint, i dont feel any improvements when using this loras.. i dont like use merged models ill rather have them as individual loras, also i use the 45gb dev undistilled model, the one needed for training mostly..

    dalvogalboMay 26, 2026
    CivitAI

    i'm using it with 10 Eros, motion is great but in case of the guy, his face becomes girlish, is it just me or happens with any of you as well?

    K3NK
    Author
    May 26, 2026

    probably coz the training has mostly female faces , im cooking a v3 that seems to work better with faces, but still no male faces in the dataset...

    dalvogalboMay 27, 2026

    @K3NK that makes sense, but thanks a lot for your efforts! i hope v3 will fix it! love your other LORAs too! great stuff.

    koalabearrMay 27, 2026· 8 reactions
    CivitAI

    I think I've solved the WAN2GP problem. The solution is a bit long, and I arrived at it using artificial intelligence. It seems like a long solution, but it works 100% in my tests. @K3NK I hope this helps you reach more people with your wonderful Loras.

    (This issue was analyzed and resolved using Antigravity, the AI coding assistant designed by Google DeepMind.)

    ⚠️ How to fix AssertionError / loading crash in Wan2GP (MMGP) with this LoRA

    If you are using this LoRA in the Wan2GP (WanGP) interface and getting an AssertionError at mmgp/offload.py line 1553, here is why it happens and how to fix it:

    Why it happens

    This LoRA is merged for ComfyUI and contains keys targeting non-transformer layers (like text/audio connector layers: embeddings_connector, adaln_single, etc.). Wan2GP's MMGP loader only registers LoRA support for the main transformer blocks, so it throws an AssertionError and crashes when it encounters these extra ComfyUI keys.

    How to Fix

    We can edit the MMGP script to log a warning and safely skip those minor helper layers instead of crashing. The main LoRA weights (99%+ of the LoRA effect) are in the transformer blocks anyway, so it will still work perfectly.

    1-Go to your Wan2GP folder and open this file in a text editor (Notepad, VS Code, etc.): C:\ai\pinokio\api\wan.git\app\env\lib\site-packages\mmgp\offload.py (Adjust path if your installation directory is different)

    2-Locate the following lines around line 1553:

    loras_module_data = loras_model_data.get(module, None) assert loras_module_data is not None loras_adapter_data, meta = _ensure_adapter_meta(loras_module_data, adapter_name)

    3-Replace them with this code block:

    loras_module_data = loras_model_data.get(module, None) if loras_module_data is None: if not hasattr(model, "_lora_warned_modules"): model._lora_warned_modules = set() if module_name not in model._lora_warned_modules: model._lora_warned_modules.add(module_name) print(f"[WARN] Module '{module_name}' is targeted by LoRA but does not support/declare LoRA. Skipping this module.") continue loras_adapter_data, meta = _ensure_adapter_meta(loras_module_data, adapter_name)

    4-Save the file and restart your Wan2GP interface. The LoRA will now load and generate without crashing!


    jazzyreynard285May 28, 2026

    Wan2GP now supports these "exotic" loras natively, from current update.

    koalabearrMay 28, 2026· 1 reaction

    @jazzyreynard285 Oh my god. I spent so much time on this lol

    dillion1920May 30, 2026

    Good buddy, i used it and it was glorious.

    If it has been fixed, so be it.

    Well done!

    FromWoodMay 30, 2026

    @jazzyreynard285 I updated to mmgp-3.7.6, the error was still there and I still had to use the workaround

    koalabearrJun 1, 2026

    @FromWood The solution I mentioned in the comment works, but you don't need it. I uninstalled my Wan2gp setup (to be absolutely sure) and reinstalled it with Pinoki. Now, this LoRA works in Wan2gp without any modifications.
    @dillion1920 That's great if it helped. You're welcome.

    FromWoodJun 1, 2026

    @koalabearr I don't know about Pinoki, I will try, thanks

    mangho9123389May 28, 2026· 4 reactions
    CivitAI

    The Asian is turning into a Westerner.

    K3NK
    Author
    Jun 1, 2026

    V3 should handle faces better, kid of xD