A custom node for ComfyUI that facilitates the conversion of the Diffusion model or CLIP model to fp8 format, enhancing efficiency in processing. This tool does not support VAE conversion but allows users to utilize pre-compiled safetensor files without needing prior separation of model components.
- Enables direct conversion of Diffusion and CLIP models to fp8, streamlining workflow.
- Reduces file sizes significantly, improving storage efficiency and potentially speeding up processing times.
- Compatible with existing ComfyUI workflows, ensuring seamless integration without generating errors during image generation.
Context
This tool is a specialized custom node designed for ComfyUI, focusing on the conversion of model components—specifically the Diffusion and CLIP models—to the fp8 format. Its primary aim is to simplify the conversion process and enhance the usability of model files within the ComfyUI framework.
Key Features & Benefits
The custom node allows users to convert models directly to fp8 format without the need to manually separate the components such as unet, clip, or vae. This feature significantly reduces the complexity of the conversion process, making it accessible for users who may not have extensive technical knowledge. Additionally, the ability to work with safetensor files provided by ComfyUI eliminates the hassle of preparing individual model parts.
Advanced Functionalities
While the tool does not support VAE conversion, it successfully converts large model files, such as those from SDXL, Auraflow, and HunyuanDIT, into the fp8 format. This conversion not only compresses file sizes but also maintains compatibility with standard image generation workflows, allowing users to leverage the benefits of fp8 without encountering errors.
Practical Benefits
By utilizing this custom node, users can optimize their workflows in ComfyUI, achieving greater control over model file sizes and processing efficiency. The reduced file sizes—such as the conversion of the SDXL model from 6.5GB to 3.4GB—demonstrate a significant improvement in storage management, which can lead to faster loading times and overall enhanced performance in generating images.
Credits/Acknowledgments
The project references an original script from an external source, which is publicly available without licensing restrictions. Any concerns regarding the use of this script can be addressed by contacting the authors of the repository.