Sulphur 2 — Free Open-Weights AI Video Generator

What

Sulphur 2 is an open-weights AI video model fine-tuned from LTX 2.3 for text-to-video and image-to-video generation. It appears aimed at creators, marketers, storyboard artists, and technical users who want better motion quality than base LTX 2.3, with options to run it in a browser or locally in ComfyUI. Its positioning is a more open and workflow-compatible alternative to proprietary AI video tools, especially for users who value local control and fewer refusals on legal content.

Features

• Text-to-video and image-to-video generation: Supports both prompt-based video creation and animation from a still image, covering common ideation and production workflows.
• Fine-tuned motion quality: Trained on 125,000 real-world clips to reduce limb deformation, camera jitter, and temporal instability compared with base LTX 2.3.
• Browser-based use without local hardware: Offers server-side generation with no GPU or installation required, which lowers setup barriers for non-technical users.
• Local ComfyUI deployment: Ships with open weights, four ComfyUI workflow JSONs, and compatibility with existing LTX 2.3 setups, which simplifies adoption for users already running that stack.
• Prompt enhancement and format support: Includes a bundled Qwen 3.5 9B prompt enhancer and supports up to 1080p MP4 output in 16:9, 9:16, and 1:1 formats.

Helpful Tips

• Choose weights based on VRAM and stage of work: FP8 is the practical starting point for 16 GB VRAM setups, BF16 is positioned for highest quality, and the distill variant appears better suited to fast iteration.
• Test short, lower-resolution renders first: The page itself suggests validating motion at 720p and 5 seconds before re-rendering at 1080p for final output.
• Assess content-policy fit carefully: The product emphasizes fewer refusals on legal content, which may matter for medical, historical, or mature cinematic use cases, but teams should still review internal safety requirements.
• Verify workflow needs beyond ComfyUI: The page clearly supports ComfyUI and Hugging Face downloads, but it does not provide evidence for broader native tooling or pipeline integrations.