Sulphur-2-base

Updated
24.08.2026
Visuals
Run

Sulphur 2 base is an uncensored video model built on LTX 2.3 that handles text-to-video and image-to-video natively, with ComfyUI workflows included.

At a glance

  • License: LTX-2 Community License, inherited from the Lightricks base
  • Base model: Lightricks LTX 2.3
  • Modalities: text-to-video and image-to-video, natively
  • Files: fp8mixed 29.2 GB, bf16 46.1 GB, distill LoRA 10.3 GB
  • Extras: bundled prompt enhancer (GGUF + mmproj) and ComfyUI workflows

What is Sulphur-2-base?

Sulphur 2 is an uncensored video generation model from SulphurAI, built on Lightricks LTX 2.3. It supports both text-to-video and image-to-video natively, along with all the other LTX 2.3 formats. It is a community model: the card credits testers, mergers and two anonymous funders, and points to a recommended image-to-video merge by TenStrip for that workload.

SpecificationSulphur-2-base
Base modelLightricks LTX 2.3
ModalitiesText-to-video and image-to-video
Checkpointsdev bf16, dev fp8mixed, distil bf16
ExtrasDistill LoRA rank 768, prompt enhancer, ComfyUI workflows
Release dateMay 2026
LicenseLTX-2 Community License

How Sulphur 2 is set up

The vendor's recommended start is one of the dev checkpoints, fp8mixed or bf16, plus the provided distill LoRA for faster generation. The two routes do not stack: use the LoRA on a dev checkpoint or use the full distil checkpoint, not both at once.

The repository also ships a prompt enhancer as a GGUF file with an mmproj projector: a small local vision-language model that turns a short prompt, optionally with an image, into a detailed video prompt. The card describes running it in a local LLM app with no system prompt, then feeding the output to the video model.

Sulphur 2 hardware requirements

This is big-GPU territory. The file sizes from the repository:

FileSize
sulphur_dev_fp8mixed.safetensors29.2 GB
sulphur_dev_bf16.safetensors46.1 GB
sulphur_distil_bf16.safetensors46.1 GB
sulphur_lora_rank_768.safetensors10.3 GB

Plan for a high-VRAM GPU or a large unified-memory machine, with fp8mixed as the practical entry point. Community GGUF quantizations exist for tighter setups.

How to run Sulphur 2 locally

Sulphur 2 runs through ComfyUI. Download a dev checkpoint and the distill LoRA from SulphurAI/Sulphur-2-base, then load one of the workflows shipped in the repository. Official standalone inference code is announced as coming soon on the model card.

For language models to run alongside it, see the full catalog of local models.

Sulphur-2-base license

The weights are distributed under the LTX-2 Community License inherited from the Lightricks base model. Read the license file in the repository before any commercial use, since the terms are set by the base model's license rather than a standard open-source one.

Get the weights from Hugging Face

huggingface-cli download SulphurAI/Sulphur-2-base sulphur_dev_fp8mixed.safetensors
huggingface-cli download SulphurAI/Sulphur-2-base sulphur_lora_rank_768.safetensors
# Sulphur 2 runs through ComfyUI; the repository ships ready workflows.
# 1. Download sulphur_dev_fp8mixed.safetensors (29.2 GB) or sulphur_dev_bf16 (46.1 GB)
# 2. Add the distill LoRA for faster generation (do not combine it with the distil checkpoint)
# 3. Load a workflow from the workflows/ folder in the repo
# Official inference code is announced as coming soon on the model card.
Desktop



macOS
(M1 or better)
Download
Windows
(x64)
Download
Linux
(x86_64)
Download
Frequently asked questions
Sulphur 2 is an uncensored video generation model from SulphurAI, built on Lightricks LTX 2.3. It supports both text-to-video and image-to-video natively, along with the other LTX 2.3 formats, and ships with ComfyUI workflows in the repository.
The fp8mixed checkpoint is 29.2 GB and the bf16 checkpoint is 46.1 GB, plus an optional 10.3 GB distill LoRA for faster generation. That puts it on high-VRAM GPUs or large unified-memory machines; the vendor recommends starting with one of the dev checkpoints plus the distill LoRA.
The weights are free to download and are distributed under the LTX-2 Community License it inherits from the Lightricks base model. Read the license file in the repository before any commercial use.
The repository ships a prompt enhancer as a GGUF file with an mmproj projector: a small local LLM that expands your short prompt, and optionally an image, into a detailed video prompt. The vendor describes loading it in a local LLM app, sending your text, and using the output as the generation prompt.