Furkan Gözükara
patreon
Ultra Advanced Wan 2.1 App Updates & Famous Squish Effect to Generate Squishing Videos Locally
🕑 Added 2025-03-10 23:25:53 +0000 UTCTutorial Video Link : https://www.youtube.com/watch?v=ueMrzmbdWBg
Squish Effect LoRA arrived to Wan 2.1. Wan 2.1 is the truly State of the Art (SOTA) Open Source video generation model that supports Text to Video (T2V), Video to Video (V2V) and Image to Video (I2V). Now our ultra advanced 1-Click Gradio application supports LoRAs and today I will show you all the new developments to our Wan 2.1 all in one video generation Gradio App. We have added so many new features since the original Wan 2.1 step by step tutorial and we continue to improve our App on a daily bases with amazing updates.
🔗 Full Instructions, Configs, Installers, Information and Links Shared Post (the one used in the tutorial) ⤵️
▶️ https://www.patreon.com/posts/click-to-open-post-used-in-tutorial-123105403
🔗 Original Main Tutorial Video
🔗 Wan 2.1 AI Video Model: Ultimate Step-by-Step Tutorial for Windows & Affordable Private Cloud Setup ⤵️
▶️ https://youtu.be/hnAhveNy-8s
🔗 SECourses Official Discord 9500+ Members ⤵️
▶️ https://discord.com/servers/software-engineering-courses-secourses-772774097734074388
🔗 Stable Diffusion, FLUX, Generative AI Tutorials and Resources GitHub ⤵️
▶️ https://github.com/FurkanGozukara/Stable-Diffusion
🔗 SECourses Official Reddit - Stay Subscribed To Learn All The News and More ⤵️
▶️ https://www.reddit.com/r/SECourses/
🔗 MSI RTX 5090 TRIO FurMark Benchmarking + Overclocking + Noise Testing and Comparing with RTX 3090 TI ⤵️
▶️ https://youtu.be/uV3oqdILOmA
🔗 RTX 5090 Tested Against FLUX DEV, SD 3.5 Large, SD 3.5 Medium, SDXL, SD 1.5, AMD 9950X + RTX 3090 TI ⤵️
▶️ https://youtu.be/jHlGzaDLkto
If you want to have Squish it: AI Squish Video Art locally for free forever, our app and Squish LoRA and Wan 2.1 is all you need. Watch this tutorial to learn all. Moreover this tutorial will show you majority of the newest features we have implemented with non-stop working for 10 days.
Hopefully many more updates coming soon.
Video Chapters
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
Comments
Furkan Gözükara
yes hopefully we will
cool1
Thanks. Will there still be a UI/app (like gradio) that you could post about for Lora training for Wan 2.1 that we can use with windows (that will work with 24GB GPUs) and if so will we be able to train it on images and video?
Furkan Gözükara
awesome
Thomas Reichert
Thank you - I had all the tools installed, so I think it was the triton cache! In any case, got it working! You're the best!
Furkan Gözükara
i just made an update for fixing RAM issue. please use latest zip file and read changelogs
Furkan Gözükara
i just made an update for fixing RAM issue. please use latest zip file and read changelogs
Furkan Gözükara
i just made an update for fixing RAM issue. please use latest zip file and read changelogs
Nedo Braun
same here, i waited 10mins before it startet, i was using the 480p modell and the 32GB VRam setup (running the rtx5090 plus i have 128GB Ram) . But i did just one test before going to work, will test later at home....
Furkan Gözükara
Triton just made an update today. That can fix issue. You are missing either of these CUDA (12.4+), C++ tools or MSVC. delete venv and reinstall today and see if it fixes. Also run clear triton cache file it can also fix
Thomas Reichert
https://prnt.sc/ZMv9bgPXuzuM Followed your python, CUDA, etc install guide to a T. Installed with installer. Get this error upon running. Any thoughts?
Furkan Gözükara
awesome. about lora i made an issue topic : https://github.com/modelscope/DiffSynth-Studio/issues/439 you can make a reply. i have 96 gb RAM so it is super fast for me but sadly 64 GB being insufficient - they need to optimize
puk
I have the same lora problem. 64 GB RAM and RTX 4090. Version 28 with new installation and new Modell Download. BTW i find out that the setting 32 GB RAM work verry well on my 4090, 13~14 minutes for i2v with the 480 Modell and a resolution by 416x608. It use 23,3 GB Vram. With 24 GB setting only 17,3 GB is in use and take 16~17 minutes.
Morgan Old
Right
Furkan Gözükara
it looks good. so lora makes it to start 15 minutes and without lora it starts immediately right?
Morgan Old
This is a My Virtual Memory https://prnt.sc/ZhUnpvT6-Cb8 , need change it ?
Morgan Old
64GB DDR5
Morgan Old
RAM 64GB, tried after restarting the computer, same situation
Furkan Gözükara
thank you so much. not an issue at all i always want feedback to improve more
Furkan Gözükara
wow this is weird. it happens instantly for me. how much RAM you have? how much virtual RAM? did you try after restart of computer?
Morgan Old
I don't want it to be such an impression that I'm not happy, this is a wonderful assembly, and I like everything, I'm sharing my results!
Morgan Old
CMD Screenshot https://prnt.sc/uAoCEjKOVZDe
Morgan Old
I want to say that when I generate a video using LoRa, I have to wait 15 minutes or more to start generating, but when I generate without LoRa, the generation starts immediately.
Morgan Old
In this moment I testing other LoRa and this is my stat now https://prnt.sc/rOo89BJZKh1j
Furkan Gözükara
50 minutes is sadly normal for 720x1280 px or 1280x720 px. this model is super heavy. once we hopefully get teacache it will be hopefully half duration. also generation should start much faster for 480p model. perhaps your RAM was busy?
Morgan Old
Thanks for answering! I have a video card RTX 4090, and with the 720 model it generates 5 seconds in 50 minutes, as I understand in your build there is a triton. I want to understand that 50 minutes is also not normal? I tried squish effect and on 480 you have to wait about 15 minutes for it to start generating
Furkan Gözükara
no not normal. try 480p model it uses lesser VRAM and faster
Morgan Old
Today I tested the Squish Effect as well. After starting it, I waited for 15 minutes, but there was no indication in the terminal that the generation process had begun. There were also no error messages. Is this normal when generating Squish Effect videos? In cmd last is this "Loading LoRA models from file: LoRAs\squish_18.safetensors"
