Wan2.2-Fun-5B-Control

Wan2.2-Fun-5B-Control

Steers generated video with a control condition such as Canny, Depth, Pose or MLSD, or with a trajectory. For video creators and researchers working with the 5B Wan2.2 base in the videox_fun library.

Made by alibaba-pai on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: A video control model that steers the clip it generates with a condition such as Canny edges, Depth, Pose or MLSD, or with a trajectory. It belongs to the Wan2.2-Fun family and is built on Wan-AI/Wan2.2-TI2V-5B. The weights are Safetensors files of 23.0 GB, trained at 121 frames and 24 frames per second. Hugging Face lists it as an image to video model for the videox_fun library.

Who it suits: Creators and researchers who want to direct how a generated video moves, instead of leaving all the motion to a text prompt.

Industry | Who uses it | How they use it | Example
Video production | Video editors | Guide a generated clip with a pose or depth video | Make a still character follow a recorded motion
Advertising | Creative teams | Animate a reference image under a chosen control condition | Turn one picture into a short moving clip with set motion
Animation | Animators | Steer movement along a trajectory | Move an object along a planned path
Research | AI video researchers | Compare Canny, Depth, Pose and MLSD conditions on a 5B base | Check how each condition changes the same scene

You can skip it if you only want video from a text prompt or from first and last frames, because the sibling Wan2.2-Fun-5B-InP covers that.

Where to use it: It runs through the videox_fun library, and the card points to Hugging Face and ModelScope for the weights.

• The videox_fun library, which the Hugging Face listing names for this model
• Windows, macOS or Linux with Python, to download the files
• Hugging Face or ModelScope, where the card links the weights

Use case | Who | What they get back
Pose-driven clips | Video creators | A video that follows the motion of a pose control video
Edge or depth guided clips | Designers | A video that keeps the structure of Canny edges or depth maps
Trajectory control | Animators | A video whose movement follows a chosen path

How to set it up: Get the files first, then follow the card's own run steps. The card has Quick start and How to use sections for that part.

• Python, for the Hugging Face download commands
• About 23.0 GB of free disk space for the weights
• The videox_fun code project by aigc-apps, which the card links

1. Install the Hugging Face tools with pip.
2. Download the model files to your computer.
3. Open the card's Quick start and How to use sections and run the weights in videox_fun.

pip install -U huggingface_hub
hf download alibaba-pai/Wan2.2-Fun-5B-Control

How to use it: You pick a control condition, prepare the matching control input, and let the model produce a video that follows it. The card says it supports multilingual prediction, with English and Chinese listed. The exact run commands sit in the card's How to use section.

Tool | What it does
Canny | Control condition based on edge maps
Depth | Control condition based on depth maps
Pose | Control condition based on body pose
MLSD | Control condition based on straight line segments
Trajectory control | Steers movement along a chosen path

Example: Pair a reference image with a pose control video, the combination the card shows for its 14B Control sibling. You get back a video in which the pictured subject moves the way the pose clip does.

For newcomers and experts: A control model lets you steer a video with more than words. You give it a picture or a motion clip, and it follows that guidance. A first small try is to run the two download commands, then read the card's Quick start section before you generate anything.

For experts, this is the 5B Control variant, built on Wan-AI/Wan2.2-TI2V-5B and trained at 121 frames and 24 frames per second. The 14B Control siblings were trained at 81 frames and 16 frames per second with 512, 768 and 1024 resolutions. The trade is size against capacity, since the 5B files take 23.0 GB and the 14B files take 64.0 GB.

Choose a control condition
Prepare a control video or trajectory
Run the weights in videox_fun
Collect the generated video

Good to know: The card lists six weight sets in all, and their storage sizes compare as follows.

Title: Storage size of each Wan2.2-Fun weight set
Wan2.2-Fun-A14B-InP: 64.0 GB
Wan2.2-Fun-A14B-Control: 64.0 GB
Wan2.2-Fun-A14B-Control-Camera: 64.0 GB
Wan2.2-Fun-5B-InP: 23.0 GB
Wan2.2-Fun-5B-Control: 23.0 GB
Wan2.2-Fun-5B-Control-Camera: 23.0 GB

• Family: InP, Control and Control-Camera each come at 14B and 5B sizes.
• Other versions: Hugging Face shows one quantized and five fine-tuned versions made by other people.
• License: Apache-2.0.
• Cost to run: none named, no API key, account or paid service, only your own computer.
• Data it reaches: it downloads the weights from Hugging Face and reads the images and videos you give it.
• Not for: the card names no limits in the text given here.

On Hugging Face, 13 November 2026

• Task: image to video
• Library: videox_fun
• Formats: Safetensors
• Downloads in the last 30 days: 25,525
• Likes: 22
• License: apache-2.0
• Languages: en, zh
• Based on: Wan-AI/Wan2.2-TI2V-5B
• Quantized versions: 1
• Fine-tunes: 5
• Tags: videox_fun, diffusers, safetensors, ti2v, video, video-generation, wan2.2, image-to-video

Links

• Model: https://huggingface.co/alibaba-pai/Wan2.2-Fun-5B-Control
• Files: https://huggingface.co/alibaba-pai/Wan2.2-Fun-5B-Control/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download alibaba-pai/Wan2.2-Fun-5B-Control

Then follow the usage in its model card. Nothing in this kit runs by itself.