Sulphur 3
an uncensored video generation model based on MiniMax H3
2 min readupdated
Sulphur 3 is an uncensored video generation model built on the MiniMax H3 architecture. It lets you generate video using standard MiniMax H3 workflows, while also offering lower VRAM requirements thanks to quantized versions in the quantized/ folder. This model is useful for content creators and researchers interested in video generation without the strict restrictions of the base engine.
Key features
- Based on the MiniMax H3 architecture with support for standard launching via
s3.safetensors - Availability of versions without RL lora (
s3_raw.safetensors) and as a standalone LoRA - Optional corrector file (
s3_corrector.safetensors) for improved generation - Includes the
S3_ComfyUI.ziparchive with custom nodes and a ready-to-use ComfyUI workflow
What you can do
- Generate video clips in ComfyUI by connecting the Sulphur 3 model weights and optional corrector
- Run the model on a GPU with lower VRAM using files from the
quantized/folder - Experiment with the raw version of the model or the standalone LoRA (
s3_raw_lora.safetensors)
Example
Download the s3.safetensors file and run it like a standard MiniMax H3 model in your environment, or use ComfyUI with the provided S3_ComfyUI.zip archive to work with the corrector.
Read more
DetailsBase model: MiniMax H3
File formats: .safetensors (including main weights, raw versions, LoRA, and optional corrector)
Hardware requirements: Lower VRAM options available in the quantized/ folder
License: MiniMax H3 Community License














Comments