SDLI (Stable Diffusion Latents in Imagefile) is a specialized tool within ComfyUI that allows users to save latent representations and their decoded images in a webp format. This functionality is particularly useful for managing and manipulating image data generated by Stable Diffusion models.
- Enables the saving of latent data alongside its decoded image, facilitating easy access and usage.
- Offers options to customize the output filename and include additional metadata for better organization.
- Supports image size reduction specifically for webp format, optimizing storage without compromising latent quality.
Context
This tool is designed to enhance the ComfyUI workflow by allowing users to save and preview images in the SDLI format, which combines both latent and decoded image data. The primary purpose is to streamline the handling of image outputs from Stable Diffusion models, making it easier for users to manage their generated content.
Key Features & Benefits
The Save SDLI Image node allows users to input latent samples and optionally a Variational Autoencoder (VAE) for enhanced image quality. Users can specify the type of latent data, customize output filenames, and add metadata through prompts, which helps in organizing and identifying files efficiently.
Advanced Functionalities
The tool includes a Preview SDLI Image node that generates a preview of the saved SDLI images in webp format. This feature allows users to quickly view their outputs in a browser without needing to open additional software, streamlining the review process.
Practical Benefits
By integrating the SDLI tool into their workflow, ComfyUI users can significantly improve their efficiency in managing image outputs. The ability to save both latent data and decoded images in a compact format enhances control over the generated content while reducing storage requirements.
Credits/Acknowledgments
The SDLI tool is based on contributions from various developers, with detailed format specifications available at the SDLI Tools repository. Additionally, users are encouraged to utilize the TAESD model for optimal results when working with VAEs.