CivArchive
    LTX 2.3 MisticNSFW AIO 10eros - v1.0
    NSFW

    The model pairs well with the OmniNFT RL lora it helps the sound a lot and improves texture quality | There should be at least 6 videos up if not then Civit is being slow with the rating system | If this model does well with everyone then there will be future versions for the new LTX model light tricks is making I think its named LTX next | This lora is my largest NSFW training run trained on 3775 videos and keeps character identity well. Trained on oral, missionary, titfuck, doggy, cowgirl etc pretty much all NSFW positions. Its been trained on both 2D and 3D, furry and standard NSFW animations. It handles these all very well but has not been created to replace all other loras. Even Sulphur 2 requires extra loras for most things and its a FFT. It was trained on 10eros 1.4 but videos are generated with 10eros 1.2 because I did not have much luck getting good gens on 1.4 with the DMD lora or normal distilled lora not really sure why but could just be a skill issue on my end. The lora has not been overtrained on the dataset so you should not have any style bleed if your doing videos that do not involve any anthro. It does well with not contaminating 2D generations with 3D look which is something I notice most loras will do to 2D since most training data is not 2D. If you train with too much 2D relative to 3D it will cause issues like frame drops and smearing, I think that has something to do with the VAE's compression not playing well with just pure 2D. But if you have a good ratio of 2D to 3D data preferably much more 3D than 2D it turns out well. Most of the training data is 3D animation with only about ~10% of the training data being 2D

    It wasn't trained on audio since most videos either had no audio or low quality audio so you may need to use the NSFW furry lora by mylo for audio only. If 10eros 1.4 works well for you then this lora will likely work even better if you use it but if not use something like 10eros 1.4.

    Has some known issues like random music and not perfect audio.

    You can make money with my models. Sell what you generate, use it in paid work, run it on paid services, merge it, sell the merge all of it, no permission needed. Credit is appreciated but never required. New releases might have a short early access window behind Buzz, then go free and free is permanent. Once a model is free, I will never re-paywall it or pull it for a paid platform, and you're welcome to mirror, archive, and rehost it so it stays free no matter what happens to any one site.

    Feel free to mirror any of my loras to HuggingFace

    Disclaimer & Terms of Use

    This model is provided "AS IS", without warranty of any kind, express or implied, including but not limited to the warranties of merchantability, fitness for a particular purpose, and noninfringement.

    IN NO EVENT SHALL MisticRain69 BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH USE OF THIS MODEL OR ANY OUTPUT GENERATED WITH IT.

    18+ only. You, the user, are solely responsible for anything you generate with this model and for ensuring your use complies with all applicable laws and the CivitAI Terms of Service.

    You may not:

    produce anything illegal, non-consensual, defamatory, or used for harassment, deception, or fraud;

    create any content depicting, or appearing to depict, minors (real or fictional) in a sexual context — no exceptions, ever;

    use this model to depict any real, identifiable person — celebrity, public figure, or private individual — in any context, SFW or NSFW. No likeness use, period.

    If you post outputs to the gallery that violate the CivitAI TOS, you'll be blocked and reported. Don't do this shit. Don't be fucking sus.

    Description

    FAQ

    Comments (25)

    ArtiOtakuJul 28, 2026
    CivitAI

    Are you talking about LTX 2.5 in your description?

    MisticRain69
    Author
    Jul 28, 2026

    Yes whatever they are going to call the new model. In the LTX discord I always see the LTX team refer to it as LTX next.

    Nk0GGryJul 28, 2026
    CivitAI

    Exciting! Is this intended to stack with your better NSFW motion lora?

    MisticRain69
    Author
    Jul 28, 2026

    Its on a case by case basis. Like with most loras sometimes it works best when stacked with a motion loras other times it works better on its own. Really depends on what your generating.

    NealPageJul 28, 2026
    CivitAI

    Are you planning to do anything about penises and bodily fluids?

    tenstripJul 28, 2026· 8 reactions
    CivitAI

    Testing this made me revisit the first pass sigmas in my WF, and I figured out how to stop the weird motion and hesitation which happened before and a lot with this lora and DMD on 1.4. These 11 step sigmas produce much better motion on first pass: 1.000, 0.968, 0.926, 0.875, 0.812, 0.741, 0.661, 0.574, 0.482, 0.241, 0.121, 0.000

    MisticRain69
    Author
    Jul 28, 2026

    Thanks! I'm not sure why it does it with the DMD on this lora. I will have to see if I can add custom sigmas into wan2gp. Its trained on the full bf16 model so its not a quantization artifact from training and all testing was done on the bf16 model as well.

    tenstripJul 28, 2026

    @MisticRain69 From what I notice the lora is actually fighting the model it seems. I tried on 3D images on 1.4 with and without and base 1.4 tends to want to do it's own pace of sexual motion and the lora starts to alter that pace, you can find a strength balance to get them to align and the output is smooth, or altering the sigma steps just lets it work it out better. Imo with that amount of clips in the dataset I would say this is a very generalized lora that is more of like a motion style, using it past 1.0 strength works as well.

    willshawn2519Jul 29, 2026

    @tenstrip how many lora strength would you recommend?

    tenstripJul 31, 2026· 1 reaction

    @MisticRain69 literally just based on this and things I've noticed; I made a new Eros version to correct that. The v1.5 is on my huggingface with a hybrid DMD lora, it basically brings back those kinds of motions from 1.2 without the issues, at least thats the attempt. This lora of course is gonna shine using that newer hybrid model and hybrid DMD lora, did plenty with it. I'll upload it here next week.

    MisticRain69
    Author
    Aug 1, 2026· 1 reaction

    @willshawn2519 Start at 1.0 strength

    willshawn2519Jul 29, 2026
    CivitAI

    Hi.
    I'd like to ask set of loras that you activate when running 10eros 1.4 (except DMD).

    1. OmniNFT RL-LoRA at what strength?
    Sorry that i'm new to this lora tbh.

    2. your Reasoning v4 at 1.0 -- am i using correctly?

    3. this NSFW lora at 1.0 -- per your recommend

    4. NSFW furry by Mylo -- at what strength to affect to audio only? I know it removes unwanted bgm, but it generated more ghosting in my experience.

    MisticRain69
    Author
    Jul 29, 2026

    For the mylo lora you need a specific node to only enable the audio from it I forgot what the name of it was. OmniNFT I run at a 1.0 strength on eros 1.2 and the previews were not using the reasoning lora for them. This NSFW lora at 1.0 strength.

    willshawn2519Jul 29, 2026

    @MisticRain69 For reasoning lora, is it correct to use v4 on 10eros 1.4?

    MisticRain69
    Author
    Jul 30, 2026· 2 reactions

    @willshawn2519 yes

    Nk0GGryJul 30, 2026· 1 reaction

    If you havent found a solution already: For the node to tweak audio weights, I use DaSiWa LTX-2 LoRA Loader node. You can adjust overall, video and audio weights separately. It even has pretty colors!

    willshawn2519Jul 31, 2026· 1 reaction

    @Nk0GGry Adapted lora loader advanced from KJnodes, thanks for advising tho.

    tazmannner379Jul 29, 2026
    CivitAI

    Training on 2d will cause more of animated on 2's style of animation which seems choppy if not applied to 2d animation. Furthermore I have a theory that the vae itself for LTX is not detailed enough to capture this type of motion in 2d thus you get the distortions. You can reduce them by giving more frames. That's why 3d and irl tend to not have as much issues. I think it wont ever be fixed in this version of ltx. Maybe a full fine tune on 2d data of the model could resolve it. Sulphur also has also been trained on a lot of 3d data it looks like, and I actually get better 2d results than with vanilla ltx model.

    DigitalGarbageJul 29, 2026· 1 reaction
    CivitAI

    Works really good with 2D. Finally some LoRA that doesn't change the style into 3D or realistic. Good job, mate.

    ElGordoAIAug 1, 2026

    for real?

    Nk0GGryJul 31, 2026· 4 reactions
    CivitAI

    For PERFECT 60fps NSFW videos with clean sound using this AIO Lora, use this Lora stack with a node to tweak audio weights:

    - MisticNSFW AIO 10Eros (this one)

    - LTX-2.3-OmniNFT-RL-Lora (huggingface)

    - Mylo's NSFW(+furry) multi purpose sex lora at 1.0 audio and 0.0 video

    YMMV, I am using DaSiWa's LTX DragonLeap V4 I2V checkpoint which is distilled, and DaSiWa's LTX workflow which uses a three-pass setup. stack extra loras with caution, start with low weights to test. happy gooning, fellow degenerates!

    OhNoMyChickensAug 1, 2026· 4 reactions
    CivitAI

    Literally like the only one who caters to the furries among the Sulphur/video gen crowd. Keep it up! Minimax H3 releases tomorrow and there's quite a few of us that will evaluate what it can do. Whether it'll be the next home for Sulphur or not. Either that, Flux3 or LTX Next. We shall see!

    So do keep an eye on that and your datasets handy. It looks very promising so far.

    MisticRain69
    Author
    Aug 1, 2026

    Also been looking forward to H3. Gotten really good furry gens on the API so far ofc not really NSFW but the anatomy and anthro features+animations have been spot on.

    OhNoMyChickensAug 1, 2026

    @MisticRain69 I saw a video a while ago featuring a Fox Girl and the animation on her hair, tail, ears and so on was spot on. So that is very encouraging. The trouble with Loras is that they often reduce prompt adherence compared to the base model, so I always try to stick to stock Sulphur Experimental whenever possible. But I tend to apply some of your Loras to furry content, which is pretty often. Hopefully H3 turns out to be amazing out of the box and further funetunes serve to add but also enhance existing knowledge.

    MisticRain69
    Author
    Aug 1, 2026

    @OhNoMyChickens I think the prompt following degradation is from loras commonly not being trained on a lot of data and being overtrained. Thats why I keep scaling my datasets up it increases generalization and prompt following massively.