This tutorial demonstrates how to use Teacache to significantly accelerate AI generations in SwarmUI with ComfyUI backend, while sharing optimal presets and detailed instructions for Wan 2.1 Text-to-Image and Text-to-Video models. Learn how to install and configure Teacache, which estimates differences between model outputs across timesteps to speed up inference while maintaining quality. Follow step-by-step guidance for both Image-to-Video and Text-to-Video generation workflows, including proper resolution settings, model selection, and parameter optimization. The tutorial covers essential configurations like frame interpolation, CFG values, and threshold settings, while explaining how Teacache works alongside technologies like Sage Attention to boost performance. Includes demonstrations of real-world speed improvements, GPU usage monitoring, and comparisons between different generation approaches.

Syllabus

0:00 Introduction: Teacache & Wan 2.1 Presets for Swarm UI
0:35 Prerequisites: Previous Tutorials & Updating Swarm UI Files
1:09 Running the Swarm UI Update Script
1:21 Importing the New Presets into Swarm UI
1:46 Enabling Advanced Options & Locating Teacache Installer
1:57 Understanding Teacache: Faster Generation, Minimal Quality Loss
2:14 Monitoring Teacache Installation Process via CMD
2:32 Teacache Installed: Preparing for Image-to-Video Generation
2:43 Applying Image-to-Video Preset & Initial Configuration
3:04 Selecting Init Image & Base Model e.g., Wan 2.1 480p
3:25 How to Download Models via Swarm UI Downloader
3:52 Choosing Specific Image-to-Video Models FP16/GGUF Q8
4:04 Setting Correct Resolution & Aspect Ratio from Model Metadata
4:25 Key Image-to-Video Settings: Model Override & Video Frames
4:42 Optimizing Video Steps 30 & CFG 6 for Teacache
5:01 Configuring Teacache Mode All & Threshold 15%
5:08 Setting Frame Interpolation 2x for 32 FPS & Duration
5:22 Starting Image-to-Video: Importance of Latest Swarm UI
5:41 Generation Started: Teacache & Step Skipping Explained
6:05 Observing Teacache in Action: Step Jumps & How It Works
6:23 Leveraging Sage Attention & ComfyUI's Automated Setup
6:38 Teacache Performance Boost: Example Speed Increase IT/s
6:51 Understanding ComfyUI Block Swapping & Monitoring GPU Usage
7:18 Image-to-Video Generation Complete: Total Time & Output
7:32 Accessing Generated Video & Output Format Options H.265
7:55 Text-to-Video: Applying Preset & Adjusting Core Settings
8:13 Configuring Text-to-Video Parameters: Steps 30, FPS, Format
8:27 Selecting Text-to-Video Model GGUF Q8 & Setting Resolution
8:45 Advanced Settings: UniPC Sampler, Sigma Shift 8, CFG Impact
9:03 Enabling Teacache 15% for Text-to-Video
9:15 Starting HD Text-to-Video Generation GGUF Q8 Model
9:36 Understanding Performance: HD Resolution & Frame Count Impact
9:54 Text-to-Video Complete: Time Taken & Teacache Speedup
10:06 Downloading & Reviewing the Full HD Text-to-Video Result
10:19 Comparing Prompt Effectiveness: Image-to-Video vs. Text-to-Video
10:30 Conclusion: Future Presets & Power of Swarm UI with ComfyUI