ComfyUI and Wan 2.1 | DW Innovation Blog

It's Time To Have Fun With AI Videos

Isn't this old news? Well, the answer is both yes and no. Sure, the technology has been around for a couple of years now, but what makes this truly unique is that you can run it locally from a desktop computer, which also means subscription-free. It does, however, require a powerful GPU to generate good results, and in my case, I am using an NVIDIA 4070 TI Super with 16GB RAM.

16GB

GPU RAM

22 min

Render Time (4s clip)

15fps

Frame Rate

The Benefits of Running AI Locally

I know many YouTube influencers are promoting online AI tools left and right, but just remember when you upload or type in your data on these sites, they collect that data for further analysis. This is both good and bad. The good part is that collecting data from many users can help developers build better AI tools, the bad news is that some may use AI tools without thinking about the data they are actually sharing with the tool itself. There is little to no ethics when it comes to this, so you really need to think twice before uploading your images or your documents to a server, and especially data that is protected by local laws, work-related, or sensitive information.

Data Privacy Warning

If you are in doubt about whether you should use AI for a specific task, you may want to ask a professional, a coworker, friend, or family member. Don't blindly trust that the data you share is kept for safe keeping only.

That's why running AI locally is always the best choice. Though you may not be able to compete on the level of speed, you do have the safety of controlling your own data, which also gives you a whole new level of freedom.

All of the examples I will be sharing with you here are created locally. I had to downscale them to make them less heavy to load on my website, and I also had to change the frame rate to 15fps. The render time for a four-second 480p (16:9) clip is about 22 minutes for my hardware setup. Though this doesn't give us the best quality, it does show us what to expect in the future.

The Logo Challenge

Sometimes when you have to develop a new website, you may have to create a new logo, and in this case, I had a little fun by giving a few celebrities the challenge to create the logo for me.

I used Flux Dev to create the images for this video. I downloaded the following LoRAs from Civitai: Gandalf, Frodo, Indiana Jones. This video was, of course, created for fun, and though it isn't perfect, it does give a clear image of where video production is heading.

The Short Story

A four-second clip doesn't give you much to work with, however, you can save the last frame from your previous render and create an additional four-second clip which you can stitch together with your previous clip to give the scene a longer duration. Creating an eight-second story script is still a bit of a challenge, and I am not sure how much success you'll have with that, but it was enough for me to have a little fun.

I used Flux Dev to create the image for this video. Rather than doing a video render of the whole image, I decided to cut out the man using Photoshop and layer him on top of a green background RGB (0,255,0). This gave me the option to green screen him back into the scene in After Effects without having to worry about rendering the background scene.

Give Your Online Visitors a Unique Experience

If you're selling products on an online platform, the best you can do to give yourself a competitive advantage is to make the customer experience as convincing as possible. Remember, price is not everything, and never underestimate the power of a good customer journey, as there is no sales staff to lead the customer from A to B. Now you can use AI to make yourself and your products stand out, and below I show some basic examples. However, should you decide to use AI for commercial purposes, I would highly recommend rendering your inputs at a high quality (720p), despite the time it may take to render it. Though you can use an upscaler to bring back the sharpness, it doesn't compensate for the loss of details in your render.

Planning

Outline exactly what you want to achieve before starting the rendering process

Quality

Render at 720p minimum for commercial use despite longer render times

Prompt Engineering

Use clear, detailed prompts to minimize trial and error

If you decide to render at a high quality, make sure you outline exactly what you want to achieve to minimize having to do the whole rendering over and over again. Remember, AI is very random if it doesn't have clear and straightforward instructions. If you want to learn more about prompt engineering, I would highly recommend reading the article "The Art of Prompt Engineering". This will help you understand why it's important to get those details right before you execute a lengthy task.

I used Flux Dev to create the image for this video. Rather than doing a video render of the whole image, I decided to cut out the leg and shoe using Photoshop and layer it on top of a green background RGB (0,255,0). This gave me the option to green screen the whole thing back into the scene in After Effects without having to worry about rendering the background scene or having to rotoscope my scene.

My Workflow Process for Commercial AI Videos

1

Create Base Images with Flux Dev

Generate high-quality initial images that will form the foundation of your video

2

Separate Elements in Photoshop

Isolate key elements against green screens for easier post-processing

3

Generate AI Animation

Use Wan 2.1 to animate your key elements with natural motion

4

Composite in After Effects

Combine all elements, add transitions, and finalize your commercial

I used Flux Dev to create two images for this video: one of the young girl and one of the shampoo bottle. The young girl was rendered in one piece, so I could add the shampoo bottle in After Effects and tweak it with some small animations such as position and rotation. An alpha splash stock video was added just to give it that commercial look.

Showcasing Your Own Artwork

I decided to take a few steps back to see what I could dig up in my old dusty archives, and amazingly, I managed to find some of my old artwork. This was, of course, from back in the day when the manual way of doing things was the technology of the century. The technology of today is a bit more advanced and a lot more effective. Though you do not have control of every detail when you let AI do the artwork, you are still able to make some incredible art pieces. So whether you are using AI to generate your art, or you are more of a "do-it-yourself" kind of person, why not have some fun with it and make it all come to life.

There is still something satisfying about having done the artwork yourself, however, I do like how AI is opening the door for more people to express themselves through art. Art is a fantastic expression tool, where all kinds of emotions can come to life, and now with LLM models like Wan 2.1, we can now take it even a step further. I had a lot of fun making these old images come to life, and I can't help but think what is next in terms of AI? Wan 2.1 definitely added a whole new dimension to it all.

Accessing ComfyUI and Wan 2.1

If you are not already using ComfyUI, I would highly recommend installing it. You do not have to be a tech engineer to work with ComfyUI, or a programmer for that matter. Just go to Pinokio to get the tool. It's free and will save you all the trouble of installing things manually. Of course, if you like to have control of everything yourself, I recommend visiting GitHub to either clone the code using git, or downloading and extracting the content to any folder of your choice.

Get Started with ComfyUI

Choose the installation method that works best for you to begin creating your own AI videos locally:

Easy Install via Pinokio (Recommended) GitHub Repository Get Help with Google AI Studio

With Pinokio, you can just quickly search for ComfyUI, where you can also easily install it (recommended). Once installed, you need to get comfortable with building workflows. Do not let this process scare you. The ComfyUI community is very big, and many experienced and professional users are sharing workflows or providing great and easy tutorials on how to set it up. You do not have to be a professional to make this work. A quick search for "Workflow Wan 2.1 ComfyUI" on YouTube will get you very far. While you are at it, I would also highly recommend going deeper into Flux to help you generate your own images, which you can use for your video renders.

Google AI Studio may also be of great help to get started with ComfyUI. You can share your screen with Google AI Studio. The good thing about Google AI Studio is that it works like instant support where you can actively talk to an AI assistant who can monitor your process and give you great advice when needed.

NVIDIA and CUDA

As mentioned earlier in this article, working locally with AI will require more GPU. This means a strong graphics card with at least 12 GB of RAM. Some LLM models will support 8 GB RAM, however, the tendency shows the more GPU power, the better the results of the output, so please keep that in mind. Also, NVIDIA supports CUDA, and some LLM models are built for CUDA only, so please make sure to keep that in mind as well before you start shopping for your next graphics card.

NVIDIA RTX 4070 Ti+

For serious AI video generation, you'll want at least an RTX 4070 Ti or higher with 16GB+ VRAM for optimal performance.

Minimum 12GB VRAM

While some models can run on 8GB, you'll need at least 12GB VRAM for acceptable quality and more complex scenes.

CUDA Required

Most AI video models are optimized specifically for NVIDIA's CUDA technology, making NVIDIA GPUs essential.

Local Video Rendering with Wan 2.1

It's Time To Have Fun With AI Videos