What is Sulphur 2 Base?
Sulphur 2 Base is not a chat model or a general text assistant. It is a video-generation model used to create short AI videos from either written prompts or input images. Public model descriptions position it as a fine-tuned derivative of LTX 2.3, supporting native text-to-video and image-to-video generation, along with compatibility for LTX 2.3-style workflows.
Sulphur 2 Deployment Tutorial:
1. Download the latest version of the ComfyUI client.
Simply install after downloading. If you previously installed an older version of ComfyUI, it is also recommended to upgrade to the latest version; otherwise, some models may be incompatible and unable to run properly.
2. Download the Sulphur 2 open-source, uncensored model.
Currently, this model is available in a full-fledged version and a distilled version. The full-fledged version requires at least 32GB of VRAM and is suitable for users with high-spec computers. For users with less VRAM, choose the distilled version, which is a compressed GGUF model.
Tips: If your video memory is greater than 32GB, you can choose the BF16 precision model for the highest quality output. If your video memory is 8GB or more, it is recommended to choose the FP8 precision model, which requires less video memory and is faster. Note: Model size is not equal to video memory size; you only need enough hard drive space.
The distilled version model is a gguf file and requires the ComfyUI-GGUF plugin to use! Place the model in the corresponding folder.
Sulphur 2 Image to Video Generator
Download the workflow. Restart and enter the ComfyUI client. Download LTX-2.3: Image-Raw Video from (Templates - Videos) to obtain the generated workflow. After loading, switch the default main model to the previously downloaded, uncensored Sulphur 2 model.
Sulphur 2 txt to Video Generator
Once you have the workflow, simply drag it into the ComfyUI client, enter the prompts you need in the text input field, and you can generate a video from the text!
Local deployment of Sulphur 2 FAQ
- Model version and workflow mismatch. For example, the workflow requires the dev version, but other weights are actually downloaded.
- Duplicate loading of LoRA and the complete model leads to abnormal effects or excessive VRAM usage.
- Insufficient VRAM, especially noticeable at high resolutions, long frame rates, and complex nodes.
- Outdated tool version; incompatible ComfyUI nodes, Diffusers, Transformers, or Accelerate versions.
- Missing supporting files such as VAE, text encoder, mmproj, and prompt enhancer.
- File paths or directory structures do not meet tool requirements.
- Just copying commands from the Hugging Face page without checking if they match the main video generation workflow or text components.
- When troubleshooting, follow this order: First, check if the model files are complete. Then, confirm the version needed for the workflow. Next, lower the resolution and frame rate. Finally, add LoRA, prompt enhancer, and post-processing nodes gradually. Changing only one variable at a time makes it easiest to pinpoint the problem.