How to generate SVD video with SD Forge

You can use Stable Diffusion WebUI Forge to generate Stable Video Diffusion (SVD) videos. SD Forge provides an interface to create an SVD video by performing all steps within the GUI with access to all advanced settings.

In this post, I will walk you through how to set up and run SVD on Forge to generate a video like this:

Software

Alternatives

Instead of running SVD on Forge, you can also use it on

• ComfyUI
• Google Colab
• Your local machine as a standalone application

Follow the instructions in this post for these options.

Install SVD model

You must first download the SVD XT model and put it in Forge’s svd folder.

First, you need to fill out the agreement form here.

Then, download the SVD 1.1 model.

Put it in the webui_forge_cu121_torch21 > webui > models >svd folder.

If SD Forge is running, you must close the terminal and restart completely to use the model.

Use the SVD model to generate a video

Step 1: Generate an initial image

You need an initial image of the video with 576 x 1024 pixels.

• Checkpoint model: juggernautXL_v7
• Prompt:

a woman with long hair windy and tattoos on her body, a green shirt , Clint Cearley, epic fantasy character art, cyberpunk art, fantasy art

• Sampling method: DPM++ 2M Karras
• Sampling Steps: 20
• CFG scale: 7
• Seed: -1
• Size: 576×1024

I used “windy” in the prompt so that her hair is blown up a little by the wind. That will help to add motion to the video.

Press the Send to SVD button under the image.

Step 2: Generate SVD video

Now, you should be on the SVD tab.

Make sure you see the SVD checkpoint selected.

You can leave the rest of the settings as default.

Press the Generate button.

You should get a video like this.

SVD settings

Below is a list of important parameters that control the video output.

Motion bucket id

The motion bucket id controls how much motion is in the video. A higher value means more motion. Accepts a value between 0 and 255.

FPS

The frames per second (fps) parameter controls the number of frames the model generates. Stay between 5 and 30 for optimal performance.

Augmentation level

The augmentation level is the amount of noise added to the initial image. Use it to change the initial image more or when generating videos that deviate from the default size.