Technology

Sulfur-2 Base: The NSFW Full Fine-Tune That Just Gave LTX 2.3 a “Pony-for-SDXL” Moment

|Author: Viacheslav Vasipenok|2 min read| 8
Sulfur-2 Base: The NSFW Full Fine-Tune That Just Gave LTX 2.3 a “Pony-for-SDXL” Moment

The AI video world just got its Pony Diffusion moment — and it’s called Sulfur-2 Base.

A small but extremely dedicated team has released a complete full fine-tune (not just a LoRA) of LTX 2.3 specifically built to remove every last trace of censorship and finally make the model actually good at generating high-quality NSFW content. Out of the box, LTX has always been notoriously weak when it comes to explicit anatomy and adult scenes. Sulfur-2 fixes that — aggressively.


How They Did It

Sulfur-2 Base: The NSFW Full Fine-Tune That Just Gave LTX 2.3 a “Pony-for-SDXL” MomentThe team trained the model on a massive dataset of 125,000 real video clips — each exactly 10 seconds long at 24 fps. They deliberately filtered out all anime and 2D content because it was actively hurting the realism and quality of the results. Only uncompromising, real-world hardcore material made the cut.

The data-labeling phase alone cost $700. Training ran on rented hardware and came in at roughly $8,000 just for compute, with the total project cost ending up noticeably higher once everything was tallied.

The Result

According to early testers, Sulfur-2 Base is dramatically better at NSFW than any existing LoRAs for competing models (including WAN). It handles complex anatomy, motion, and explicit scenes with a level of coherence and detail that LTX 2.3 has never shown before.

You can check the (very NSFW) examples on Civitai: https://civitai.red/models/2594061/sulphur-2-base


Where to Get It

Sulfur-2 Base: The NSFW Full Fine-Tune That Just Gave LTX 2.3 a “Pony-for-SDXL” MomentModel weights + ready-to-use ComfyUI workflow are already up on Hugging Face:  
https://huggingface.co/SulphurAI/Sulphur-2-base/tree/main

To run it, simply replace the original LTX weights in your ComfyUI setup with the Sulfur-2 files. No extra training or complicated merging required.

Also read:


What’s Next

The team has already confirmed they are not planning to fine-tune WAN — the compute cost is simply too high right now. Instead, they’re already working on Sulfur-3, the next iteration.

This is exactly the kind of no-compromise, community-driven project that pushes open-source video generation forward. LTX 2.3 just got the adult upgrade it always needed, and the bar for uncensored video models has been raised overnight.

Welcome to the post-censorship era of LTX.  
Sulfur-2 is here — and it doesn’t hold back.

Share:
0