Wan 2.2 Setup in ComfyUI | Full AI Video Workflow Tutorial (8GB VRAM Tested) - video Dailymotion

Jean

Just released: Wan 2.2 — a powerful open-source image-to-video AI model now running smoothly in ComfyUI, even on setups with just 8GB of VRAM 💻  In this step-by-step tutorial, I walk you through the complete installation and execution process, including: • Downloading and setting up Wan 2.2 GGUF models (5B & 14B) • Installing the LoRA, VAE, and UMT5 text encoder • Running official Hugging Face workflows • Generating your first cinematic AI animation from a single image • BONUS: Using Gemini CLI to fix a Python dependency automatically  📹 This guide is beginner-friendly, fully visual, and open-source. Whether you're an AI artist, developer, or tech enthusiast, you'll be able to recreate the results on your own setup.  🎬 Watch the full setup + test results on YouTube here: 🔗 https://youtu.be/7hUO6KhUsvQ  #Wan22 #ComfyUI #AItools #ImageToVideo #AIVideo #GenerativeAI #AIworkflow #OpenSourceAI #AIArt #Tutorial

Transcript

00:00Alright, let's get WAN 2.2 fully installed in ComfyUI, including all the necessary models and dependencies.

00:08First, we're on the Hugging Face page for a GGUF version of the WAN 2.2 image-to-video model.

00:15I'm scrolling through the files here.

00:17For this setup, I'm going to download two of the 14 billion parameter quantized models.

00:22I'll click the download icon for the WAN 2.2 image-to-video-high-noise-gguf file first.

00:31And now I'll grab the low-noise file as well.

00:53Now that those are downloaded, let's move them to the right place.

00:57I have my Downloads folder open, and my main ComfyUI Portable Installation folder open in another window.

01:10For these specific image-to-video models to be recognized, they need to go into the ComfyUI Models U-Net folder.

01:22So, I'll just select both files from my Downloads and drag them right into the U-Net folder.

01:44Next up, many of these advanced workflows use Allora for extra control.

01:49I'm on the page for the Light X2 Volts Image 2 Video LoRa, and I'm going to download the Rank 32 Safe Tensors file.

02:02Okay, the LoRa is downloaded.

02:05This one goes into the LoRa's folder.

02:19And we'll move that Light X2V Safe Tensors file from Downloads into the LoRa's folder.

02:26A full WAN 2.2 workflow requires a few more key components.

02:42We'll start with the text encoders.

02:44I'll follow their link to the repackaged models on Hugging Face.

02:48This page has the text encoders we need.

02:50We'll see you next time.

03:20Okay, with the text encoders downloaded, it's time to move them.

03:25These files belong in the text encoders directory.

03:28I'll navigate to that folder now.

03:30Now I'll drag the Safe Tensors file and drop it into the text encoders folder.

03:43The last essential piece we're missing is the VAE.

03:56Back on the GitHub page, click the link for the WAN 2.1 Safe Tensors model.

04:01I already have it downloaded, but if you don't, download it through the link.

04:05This file goes into the VAE folder.

04:13Okay, all our models should be in place.

04:28I've also downloaded the WAN 2.2 Safe Tensors.

04:32This file goes into the VAE folder.

04:43Now let's grab a sample workflow to test everything.

05:13Now inside the Comfy UI, the first thing I'll do is open the manager.

05:24I'm going to click Update All to make sure all the custom nodes are current.

05:32It will prompt for a restart, which will confirm.

05:43Let the browser reload.

05:56Okay, with a fresh, updated interface, I'll click Open, then open the workflow file we downloaded.

06:13If you don't have Sage Attention installed, you will get a message like this.

06:27This means we're missing a Python dependency.

06:31I'm going to bring up my Gemini CLI.

06:34I've given it the error message and access to my Comfy UI folder.

06:38It's immediately identified the missing Sage Attention Module.

06:42Gemini CLI is smart enough to search for a better solution.

06:46It's found an installation helper script on GitHub.

06:49I'll authorize it to download and run the script.

06:53It automatically downloads and installs all the correct versions of the dependencies like Triton and Torch.

07:08Make sure to copy and paste this link when needed for the CLI.

07:20I've also added all the links in the description for easier access to all the necessary files you'll need for this workflow.

07:38After the install, it found an error related to FP-16 accumulation and diagnosed a PyTorch version issue.

07:52If you get the same message, just write it to the CLI.

07:55It's now fixing it by uninstalling the conflicting version and installing the correct nightly build of PyTorch 2.7.0.

08:04This saves a ton of manual troubleshooting.

08:06All error messages should be resolved.

08:09Restart your Comfy UI one more time and everything should be good.

08:21I'm loading a different workflow now, one that uses an image of a wolf as the input.

08:26I'll then prompt for the wolf to stand on a cliff with dark clouds and lightning.

08:36A perfect animation of our wolf standing on a cliff with dark clouds and lightning.

08:43These generations take about four minutes to generate, which is pretty fast.

08:57Just to show another example, here's a workflow ready to go with a seal wearing sunglasses.

09:05The possibilities are endless now that the setup is complete.

09:11Look at the high quality of the video.

09:14Hairs and whiskers moving.

09:16Reflection on the glasses.

09:18Clothes flapping.

09:19This looks really good.

09:20And there you have it.

09:23A complete installation successfully generated videos.

09:27I really hope this tutorial was helpful for you.

09:30If you enjoyed the video, please like and share.

09:32And be sure to subscribe to my channel for more tutorials just like this one.

09:36Thanks for watching.

Wan 2.2 Setup in ComfyUI | Full AI Video Workflow Tutorial (8GB VRAM Tested)

Category

Transcript

Be the first to comment

Recommended