Sulphur 2 uncensored AI video generator | Open weights for LTX 2.3

Overview

Sulphur 2 is an uncensored AI video generator: an open-weights, community-released video model fine-tuned on LTX 2.3 with native T2V, I2V, distill LoRAs, ComfyUI workflows, and a local prompt enhancer.

Sulphur 2 is an uncensored AI video generator released as open weights by the SulphurAI community. The Sulphur 2 base model is a fine-tune of Lightricks LTX 2.3, ships with text-to-video, image-to-video, distill LoRAs, ComfyUI workflows, and a local prompt enhancer, and is designed to run on consumer GPUs without a hosted SaaS in the loop.

Sulphur 2 is not a hosted product and not a brand-new architecture. It is a community-distributed, open-weights uncensored AI video generator built on top of LTX 2.3, and it is distributed as a complete release with weights, distill LoRAs, ComfyUI workflows, and a prompt enhancer rather than a single checkpoint.

Key Features

• Sulphur 2 uncensored AI video generator
• Try the Sulphur 2 Space
• What Sulphur 2 actually is, in plain language
• What is in the Sulphur 2 release
• Entity context
• What the Sulphur 2 uncensored AI video generator can do
• Text-to-video and image-to-video
• Drop-in for existing LTX 2.3 workflows
• Realism-biased motion
• Local prompt enhancement
• Hosted demo
• The Hugging Face Space from the reference page is embedded below so visitors can test generation directly on this page.

Details

Per the contributor's public notes, training used ~125,000 video clips at 10 seconds and 24 fps, roughly 500 GB total, filtered only to exclude illegal content and 2D animation in order to protect realism.

Sulphur 2 is a derivative of Lightricks' DiT-based LTX 2.3 audio-video foundation model, and it is described as a drop-in replacement compatible with existing LTX 2.3 workflows.

sulphur_dev_bf16.safetensors and sulphur_dev_fp8mixed.safetensors form the main Sulphur 2 base checkpoints, sized at ~46 GB and ~29 GB respectively for the full and FP8-mixed dev variants.