Video generation methods
AnimateDiffAdd motion modules to SD 1.5 models. The most flexible and widely used video method. 16 to 32 frames per generation.
Stable Video DiffusionImage to video model. Provide a starting frame and SVD generates a smooth video clip with natural motion.
Wan2.1Newer model with strong motion quality. Supports longer clips and diverse motion styles. Growing community adoption.
Frame InterpolationRIFE and FILM nodes smooth out motion by generating intermediate frames. Turn choppy 8 FPS clips into smooth 24+ FPS video.
VRAM requirements for video
Video generation is more VRAM intensive than image generation because the model must process multiple frames simultaneously. AnimateDiff with SD 1.5 works on 8GB VRAM. SVD needs 12GB+. Wan2.1 and Hunyuan Video need 16 to 24GB for comfortable operation. If your GPU is limited, reduce the frame count, use lower resolution, or use Comfy Cloud for video workflows.
Getting started with video
Start with AnimateDiff since it works with familiar SD 1.5 models and has the most tutorials and community support. Install the AnimateDiff Evolved custom node, download a motion module (available on Hugging Face), and load a community workflow to see how the nodes connect. Once you understand the frame batch concept, experiment with SVD or Wan2.1 for different motion characteristics.