AIAny
AI Model2026
Icon for item

Sulphur 3

Generates video with audio from prompts using a MiniMax H3 fine-tune, with an uncensored tuning approach. Optional quantized checkpoints reduce memory demands; the weights are large and require local inference.

Introduction

Video generation models are often judged by visual quality alone, but an H3-based model that also generates audio makes the result a more complete clip. Sulphur 3’s distinctive angle is its combination of that audio-capable base with a fine-tune intended to reduce content restrictions; the trade-off is that users take on responsibility for reviewing generated output.

Key Capabilities
  • Built on MiniMax H3, so generation follows a video-and-audio path rather than producing silent visuals alone.
  • The repository offers a primary checkpoint, a raw variant without the RL LoRA, and a LoRA file. This gives experienced users options for applying or comparing the tuning rather than treating it as a single fixed checkpoint.
  • Quantized variants are available for lower-memory hardware, while the full BF16 checkpoint is about 40 GB. Even the smaller options require a substantial local setup.
  • An optional corrector and ComfyUI workflow are included, but the model can be used without them; the corrector is not a prerequisite for generation.
Who It’s For

A good fit for local video-generation users who want to experiment with an audio-capable H3 model and are comfortable managing large weights, hardware demands, and output moderation. Look elsewhere if you need a hosted inference endpoint, a lightweight turnkey application, or a model with tightly constrained content behavior. The community license and the uncensored positioning also make policy and rights review important before public or commercial use.

Information

  • Websitehuggingface.co
  • OrganizationsSulphurAI, FusionCow
  • Published date2026/10/31

More Items

Hugging Face
AI Video2026

Distills MiniMax H3 video generation into 2 or 4 steps, with an option to predict pixels directly and skip VAE decoding. Supports synchronized audio and first- or last-frame conditioning.

Hugging Face
AI Model2026

Uses a LoRA edit to reduce Inkling 2 Mini’s refusal behavior and provides GGUF builds for local use. It accepts text and images; evaluations show refusals drop, while MMLU falls by 1.5 points.

Hugging Face
AI Model2026

Runs text generation, summarization, and question answering directly on Windows devices as a 2.3B-parameter instruction-tuned model. It supports eight languages and a 32K context window, trading broad world knowledge and deep reasoning for local execution and privacy.