This 14-minute tutorial demonstrates the latest updates to the Wan 2.1 open-source video generation model, featuring the new Squish Effect LoRA for creating squishing videos locally. Learn about the enhanced 1-Click Gradio application that now supports Text to Video (T2V), Video to Video (V2V), and Image to Video (I2V) generation. Explore numerous new features including FP8/BF16 options, RIFE for FPS boost, improved configuration saving system, and LoRA support. Follow along with installation instructions, troubleshooting tips for Triton errors, interface navigation, and optimization techniques to reduce VRAM usage and generation time. The tutorial also covers finding compatible LoRAs on CivitAI, demonstrates real-world performance on an RTX 5090, and introduces the MMAudio application for audio generation. Comprehensive chapters guide you through each aspect of the application, from basic setup to advanced configuration options and practical demonstrations of the Squish effect.

Syllabus

0:00 Introduction & Wan 2.1 Updates: Overview of new features and the Squishing LoRA effect.
0:10 Squishing Video Effect LoRA & New Features: Exploring the new LoRA and updates like FP8, BF16.
0:29 Application Updates and Demonstrations: Showing "Save Prompt to File" and config saving.
0:45 RIFE for FPS Boost, FP8/BF16 Options, and LoRA Support
1:02 Improved "Save Prompt to File" & Configuration System: How saving prompts and configs works.
1:19 Configuration Saving Demonstration Example: "My Squish LoRA": A practical example of config use.
1:37 Using the Application & Installation Tutorial Link
2:05 Navigating the Update Log and Finding Installation Instructions: Step-by-step setup guide.
2:33 Downloading, Installing, and Running the Installation Script
3:11 Downloading Models and Model Selection Options: Choosing the right model for your needs 480p, 720p.
3:44 Downloading Specific Models and File Size Optimizations: How to manage download size.
4:11 Starting the Application & Troubleshooting Triton Errors: Fixing common startup issues.
4:35 Interface Overview, Model Selection, and GPU Preset Settings: Configuring the application.
5:00 "Apply Practical RIFE" for FPS Boost and Smoothness Demonstration: Before/after RIFE comparison.
5:29 RIFE Model Details, Prompt Setup, and LoRA Activation Word: Using the Squish LoRA effectively.
6:00 Selecting the LoRA in the Interface and Uploading Your Image: Preparing for image-to-video.
6:39 Parameter Verification, Kudos to the Creator, and Frame Count Explanation: Setting up the generation.
7:11 Reducing Inference Steps for Faster Generation and Monitoring GPU Usage: Optimizing speed.
7:55 Generation Time, Future Speed Improvements Teacache, and Implementation Details
8:24 Current Optimizations Triton, CUDA, Torch, Flash Attention and Comparison to ComfyUI
8:54 VRAM Usage, Quality Focus, and Finding Other Compatible LoRAs on CivitAI
9:34 CivitAI LoRA Search, Generation Progress, and Speed on RTX 5090: Real-world performance.
9:51 Main Tutorial Video Recommendation and Prompt Enhance Improvements
10:27 Continuous Application Improvements, Future Plans Hunyuan Video, and Tiled VAE
10:58 Generation Completion and Example Video Review: Assessing the final output quality
11:11 Introducing to MMAudio: Brief introduction to MMAudio application
11:32 MMAudio demo and explanation: showing example and basic prompt
12:03 Speed of MMAudio and coming tutorial: demonstration of generation speed
12:14 MMAudio multiple seed generation: Generating multiple seeds for better output
12:41 MMAudio features and showing examples
13:12 Conclusion, Resource Links, and Call to Action