Double AI Workflow Speed with Tea Cache Hunyuan, LTX, and FLUX Workflows

Have you ever wished your AI could run faster without losing quality? That’s where TeaCache comes in! With this handy node, you can cut your video and image generation time by half. Imagine finishing a 10-minute job in just 5 or 6 minutes—sounds like magic, right? And don’t worry, your final results will look just as amazing as before.

Let me show you how TeaCache works, how to install it, and why it’s a must-have for speeding up your workflow.

TeaCache is like giving your computer a turbo boost. It uses a clever trick to skip unnecessary calculations during the AI generation process, saving time while keeping quality intact.

Here’s an example:

• Without TeaCache: A video takes 10 minutes to generate.
• With TeaCache: The same video is ready in 5–6 minutes—with 0% quality loss!

It works with popular tools like HunyuanVideo, LTX-Video, and FLUX. Whether you’re creating videos or images, this node has got you covered.

Getting TeaCache up and running is super easy.

Option 1: Install with ComfyUI-Manager

1. Open ComfyUI-Manager.
2. Search for ComfyUI-TeaCache in the node list.
3. Click Install—done!

Option 2: Manual Installation

1. Go to your ComfyUI/custom_nodes/ folder.
2. Open the command prompt (type cmd in the folder’s address bar).
3. Run this command

git clone https://github.com/welltop-cn/ComfyUI-TeaCache.git

That’s it! TeaCache is ready to use.

Once installed, adding TeaCache to your workflow is a breeze. Let me break it down for you:

1. Insert the TeaCache Node: Place the TeaCache node between the Load Diffusion Model and the sampling nodes (e.g., Model Sampling SD3).
2. Set the rel_l1_thresh: This setting tells the model to skip over tiny, unimportant differences. Use these recommended values for different models:
   • HunyuanVideo: 0.15
   • LTX-Video: 0.06
   • FLUX: 0.4

If you’re not sure about the settings, start with these, and tweak them later for even faster results!

Here’s how you can use TeaCache in different scenarios:

1. HunyuanVideo: Create stunning videos at lightning speed. Just connect TeaCache and watch the magic happen.
2. LTX-Video: Pair TeaCache with the LTX model for higher-quality videos. Save the LTX model in your “Diffusion Model” folder for even better results.
3. FLUX: Generate images faster while maintaining their artistic charm.

Each workflow also supports an alternative node called WaveSpeed, so you can experiment and see which works best for you.

Think of rel_l1_thresh as a checklist for skipping unnecessary steps. When set correctly, it speeds up the process without compromising quality.

But keep this in mind:

• Lower values (like 0.06) ensure no detail is skipped, keeping quality at its peak.
• Higher values (like 0.4) prioritize speed but might lose tiny details.