The CacheDit documentation is below.

Step 1: Debugging without CacheDit. Update ComfyUI and install or update the nodes. Launch. You can choose other models -qwen Image, zImageTurbo.

Step 2: Close ComfyUI. Installing CacheDit. ComfyUI's CacheDiT (or Cache-DiT) is a special node that accelerates image and video generation with Diffusion Transformer (DiT)—based models such as Z-Image Turbo, Qwen, or Wan. It works by caching intermediate model calculations in memory so that they are not recalculated in subsequent steps or generations, resulting in up to 2x speedup without compromising quality.

Project page - GitHub - vipshop/cache-dit: 🤗 A PyTorch-native and Flexible Inference Engine with Hybrid Cache Acceleration and Parallelism for DiTs.

-- Go to the ComfyUI/custom_nodes/ folder

-- git clone https://github.com/Jasonzzt/ComfyUI-CacheDiT.git

I downloaded it directly in a Zip file. Then I unpacked it in a folder - custom_nodes

Changed the name to - ComfyUI-CacheDiT

Installing the library - cache-dit library

-- then in the ComfyUI-CacheDiT folder

-- pip install -r requirements.txt

Done, check the node's load for conflicts after startup.

This is the instruction for the portable ComfyUI build.

Step 3: start and debug the node. Now there are selected for Z - Image maximum values. Gradually lowering, look in cmd the name of the characteristics (picture in the archive). The table for this model and others in the archive is. Lowering the data of the node goes faster generation. Find the ratio of image quality and render speed.