Sulphur 2: Uncensored Open-Source Video Generation Runs Locally

By Prahlad Menon 3 min read

The open-source AI video generation space just got more interesting. Sulphur 2 is an uncensored video generation model built on top of LTX 2.3 that runs entirely on your local hardware — no cloud, no content policies, no history logs.

What Sulphur 2 Does

Sulphur 2 generates 10-second, 24fps video clips from either text prompts or reference images. It supports both text-to-video (t2v) and image-to-video (i2v) natively, along with all other LTX 2.3 generation formats.

The key differentiator: zero censorship. While commercial video AI services (Runway, Pika, Kling) enforce strict content policies, Sulphur 2 places no restrictions on what you can generate. Everything runs locally — nothing is logged, nothing is sent to external servers.

Technical Details

  • Base model: LTX 2.3
  • Output: 10 seconds at 24fps
  • Modes: Text-to-video, image-to-video, plus all LTX 2.3 formats
  • Formats available: Full bf16, fp8 mixed precision, distilled LoRA
  • Prompt enhancer: Built-in VLM (GGUF format, runs in LM Studio)
  • Workflow: ComfyUI compatible

The model ships in multiple versions. The recommended setup is the fp8 mixed or bf16 dev version paired with the provided distill LoRA. There’s also a quantized version on CivitAI for lower VRAM configurations.

Getting Started

  1. Download the model from Hugging Face (fp8mixed or bf16 dev version recommended)
  2. Grab the distill LoRA from the same repository
  3. Set up ComfyUI with LTX 2.3 support
  4. Optional: Prompt enhancer — download the GGUF and mmproj files, load in LM Studio under a Sulphur/promptenhancer folder

The prompt enhancer is a multimodal model that takes your text (and optionally an image) and rewrites it into an optimized generation prompt. No system prompt needed — just send your description.

Hardware Requirements

Since this is based on LTX 2.3, expect similar VRAM needs:

  • Minimum: 8GB VRAM (fp8 quantized)
  • Recommended: 12-16GB VRAM (bf16 full)
  • Comfortable: RTX 4090 / 24GB for fast generation

For image-to-video specifically, the community recommends TenStrip’s i2v merge which is optimized for that use case.

Why This Matters

The uncensored angle is the obvious headline, but there’s a broader point: local video generation is becoming practical. A year ago, generating coherent 10-second video clips required cloud APIs and monthly subscriptions. Now you can do it on a consumer GPU with no ongoing costs.

For researchers, artists, and developers who need unrestricted creative tools — or simply want complete privacy over their generations — Sulphur 2 fills a gap that commercial services intentionally leave open.

Privacy Considerations

Running locally means:

  • No generation history stored on external servers
  • No content review or moderation
  • No API keys or accounts required
  • Complete control over your outputs

This is particularly relevant for anyone working in sensitive creative domains (medical visualization, security research, artistic expression) where commercial content policies create unnecessary friction.

Sulphur 2 is community-funded and developed. If you find it useful, consider supporting the project on Ko-fi.