Wan is a family of generative AI models focused on video creation, giving creators the ability to generate and transform video using artificial intelligence.
From turning a still image into a moving scene to generating video from a text description, Wan has become an interesting tool for creators who want more control over locally generated AI video.
But what exactly is Wan, and what can you do with it?
Let’s take a look.
What Is Wan?
Wan is a generative AI model family developed for creating video content.
Rather than being a single application, Wan is a collection of models and technologies designed to handle different forms of video generation.
Depending on the model and workflow you’re using, Wan can be used for tasks such as:
- Text-to-video generation
- Image-to-video generation
- Video generation from reference images
- Character animation
- Cinematic scenes
- Creative visual effects
- Experimental video generation
One of the reasons Wan is interesting to local AI creators is that it can be integrated into workflows using tools such as ComfyUI.
That means you aren’t limited to a simple online interface.
You can build your own generation pipeline and combine Wan with other AI models and tools.
What Can You Create With Wan?
The easiest way to understand Wan is to forget about the technical specifications for a moment.
Think about what you want to create.
A photograph can become a short video.
A character can move.
A camera can travel through a scene.
A static environment can become an animated shot.
A simple idea can become a cinematic sequence.
For example, you could start with an image of a character standing in a city.
Your video prompt might describe:
The woman slowly turns toward the camera while her hair moves gently in the wind. The camera slowly pushes forward.
The original image provides the visual foundation.
The video model then attempts to create the movement.
This is known as image-to-video generation.
Text-to-Video
Text-to-video generation starts with a written description rather than an existing image.
You describe the scene and the model generates the video.
For example:
A cinematic shot of a woman walking through a rainy city at night, neon signs reflected in the wet pavement, realistic lighting, slow camera movement.
The model has to generate the subject, environment, lighting and motion.
This gives you a lot of creative freedom.
However, it also means that you have less control over the exact appearance of the result.
For consistent characters and specific compositions, starting with an image can often be more useful.
Image-to-Video
Image-to-video is one of the most interesting applications of Wan.
Instead of asking the model to create everything from scratch, you provide an image as the starting point.
The image might contain:
- A person
- A landscape
- A building
- A product
- A fictional character
- A cinematic scene
You then describe how you want the image to move.
For example:
The camera slowly moves closer while the subject looks toward the horizon. Gentle wind moves the hair and clothing.
The model attempts to preserve the original visual information while introducing motion.
This creates a very different workflow from traditional text-to-video generation.
Camera Movement
Camera movement can have a huge impact on the final result.
A simple image can feel much more cinematic when the camera moves.
Common instructions include:
Slow push-in
The camera gradually moves toward the subject.
Pull-back
The camera slowly moves away from the subject.
Tracking shot
The camera moves alongside a person or object.
Pan
The camera moves horizontally across the scene.
Orbit
The camera moves around the subject.
These descriptions can be combined with subject movement.
For example:
The woman slowly walks forward while the camera tracks backward, maintaining a cinematic medium shot.
The exact interpretation depends on the model and generation settings, but experimenting with camera instructions can produce very different results from the same starting image.
Character Animation
Wan can also be useful for bringing AI-generated characters to life.
A common workflow is:
Generate character image
↓
Create image-to-video generation
↓
Add movement
↓
Generate several short clips
↓
Edit clips together
This approach can be useful for creating:
- Social media videos
- Music videos
- Short films
- Character stories
- Digital art projects
- Fashion videos
- Experimental animations
Rather than trying to generate a complete film in one generation, creators can produce many short shots and combine them during editing.
Why Use Short Clips?
AI video generation works differently from traditional filming.
You don’t necessarily need one long generation.
Short clips can actually give you more creative control.
You can generate one shot of a character walking.
Another shot showing a close-up.
Another showing the environment.
And another showing the character from a different angle.
These clips can then be edited together.
This also means that if one generation doesn’t work, you only need to regenerate that particular shot.
Wan and ComfyUI
For creators who like experimenting with local AI, ComfyUI can be particularly interesting.
Instead of treating video generation as a single button, ComfyUI allows you to create a workflow consisting of multiple stages.
A simplified workflow might look like this:
Input Image
↓
↓
Video Frames
↓
Upscaling
↓
Frame Interpolation
↓
Final Video
More advanced workflows can contain considerably more steps.
You can also combine different models and techniques depending on the result you’re trying to achieve.
This flexibility is one of the biggest advantages of working locally.
Hardware Requirements
One of the first questions people usually ask about AI video models is:
How powerful does my computer need to be?
There isn’t one universal answer for Wan.
Memory requirements depend on factors such as:
- Model size
- Model format
- Quantization
- Video resolution
- Number of frames
- Generation settings
- Workflow
- Additional models loaded into memory
Different versions and model formats can therefore have very different hardware requirements.
For local users, this is an important consideration when choosing which version to download.
We’ll cover specific hardware requirements on individual model and version pages whenever the information is available.
Wan Model Versions
Wan is not a single frozen model.
The Wan ecosystem continues to evolve, with different releases and model variants designed for different purposes.
Because of that, this page focuses on Wan as a model family rather than one particular version.
When a significant new version is released, we’ll cover it separately in the AI News section.
Individual version guides can then include more specific information such as:
- Model downloads
- Recommended formats
- Hardware requirements
- Supported workflows
- New features
- Performance
- Example generations
This keeps this page useful as a general introduction to Wan while allowing individual releases to be documented separately.
Where Can You Get Wan?
Wan models and related files are commonly distributed through AI model repositories such as Hugging Face.
Because different versions and model variants can have different download requirements, always check the documentation belonging to the specific version you’re using.
When we cover a particular Wan release, we’ll provide the relevant download information and explain which files are useful for different setups.
Prompting Wan
There is no single perfect prompt for video generation.
A useful starting point is to describe three things:
1. The subject
What is happening in the scene?
A woman walking through a forest.
2. The movement
What should the subject do?
She slowly walks forward while looking around.
3. The camera
How should the camera behave?
The camera follows behind her with a slow cinematic movement.
Put together:
A woman walking through a forest, slowly looking around as she walks forward, the camera following behind her with a smooth cinematic movement.
From there, you can add details such as lighting, atmosphere and visual style.
Things That Can Go Wrong
AI video generation still has limitations.
You may encounter:
- Inconsistent faces
- Strange hands
- Deformed objects
- Unnatural movement
- Flickering details
- Changing clothing
- Unstable backgrounds
- Unexpected camera movement
These problems are part of working with generative video.
The quality of the starting image, prompt, model version, resolution and generation settings can all affect the result.
Sometimes a small change to the prompt can make a surprisingly large difference.
Wan: A Creative Tool
The most interesting thing about Wan isn’t simply that it can generate video.
It’s that it gives creators another way to experiment with visual storytelling.
You can start with an idea.
Turn that idea into an image.
Turn the image into a moving scene.
Create several different shots.
Edit them together.
And eventually turn a collection of AI-generated clips into something much larger.
That’s where AI video starts becoming more than a technical demonstration.
It becomes a creative tool.
What’s Next?
Wan is part of a rapidly developing AI video landscape.
New versions, model variants and techniques will continue to appear.
We’ll cover those developments in AI News, while this page will remain the general introduction to the Wan model family.
For specific releases, look for our individual Wan articles where we’ll dive into the actual model, downloads, workflows, hardware requirements and results.
Want to see what Wan can actually create?
That’s where the experiments begin.
— Aiden
