Tiled Diffusion VAE for ComfyUI: A Step-by-Step Guide

Oct 24, 2020 | Data Science

Have you ever found yourself needing to upscale large images but facing the challenge of limited VRAM? Fear not! The Tiled Diffusion VAE for ComfyUI is here to rescue your digital artistry. This extension employs various sophisticated techniques to enable efficient scaling while retaining image quality. Let’s dive into how to get started!

What is Tiled Diffusion?

Tiled Diffusion is like a master chef using a variety of cooking techniques to prepare a gourmet meal. Instead of cooking everything in one pot, the chef divides ingredients into smaller pots (tiles) to manage the cooking better without sacrificing flavor (image quality). In this case, the Tiled Diffusion VAE allows for upscaling large images efficiently by processing them in segments.

Key Features of Tiled Diffusion

  • Compatibility with various models: SD1.x, SD2.x, SDXL, and SD3
  • Support for ControlNet
  • Ability to upscale images using the Img2img feature
  • Option for ultra-large image generation

Installation Guide

To get the Tiled Diffusion VAE up and running, follow these steps:

  1. Download the SD-WebUI extension.
  2. Follow the installation instructions provided in the repository.
  3. Verify that you have the necessary VRAM and dependencies installed.

Configuring Tiled Diffusion

Now that you have the extension installed, it’s time to adjust the configuration:

  • Set tile_overlap: Start with a value of 0 to visualize the seams between tiles. Once you understand how your image segments, tweak this value to minimize seams.
  • Adjust tile_batch_size: Increase this for faster processing if your system can handle it. Remember, more tiles can mean more information to juggle!
  • Color correction: If you notice color inaccuracies, utilize the colorfix node found here to rectify your images.

Understanding the Options

When configuring your Tiled Diffusion, you will encounter several options:


method: Tiling strategy
tile_width: Width of each tile
tile_height: Height of each tile
tile_overlap: Overlap between tiles
tile_batch_size: Number of tiles processed in a batch

Tile Arrangement Specification

The arrangement of tiles is crucial for achieving optimal results. You can define the arrangement using a Math Expression node in ComfyUI:


C = number of columns you want
R = number of rows you want
tile_width = pixel width of input image / C
tile_height = pixel height of input image / R

Troubleshooting Tips

If you encounter issues or unexpected behaviors, here are some troubleshooting ideas:

  • Double-check the dimensions set for the tiles to ensure they match your image requirements.
  • Experiment with different values for tile_batch_size to find an optimal speed.
  • If image colors appear off, remember to run the colorfix node for correction.

For more insights, updates, or to collaborate on AI development projects, stay connected with fxis.ai.

Final Thoughts

At fxis.ai, we believe that such advancements are crucial for the future of AI, as they enable more comprehensive and effective solutions. Our team is continually exploring new methodologies to push the envelope in artificial intelligence, ensuring that our clients benefit from the latest technological innovations. With Tiled Diffusion VAE, you can upscale your images efficiently and beautifully. Dive in and start experimenting today!

Stay Informed with the Newest F(x) Insights and Blogs

Tech News and Blog Highlights, Straight to Your Inbox