ComfyUI AnimateDiff tutorial: image to MP4 setup

This tutorial builds an image animation workflow in ComfyUI using AnimateDiff. The presenter uses the portable PC version and requires Comfy Manager. CyberRealistic SD1.5 is the base model, which the presenter says works well on PCs with low VRAM; no minimum hardware specification is given.

Setup includes downloading the AnimateDiff motion checkpoint, installing AnimateDiff Evolved and Video Helper Suite through the manager, and restarting ComfyUI. The workflow starts with the default graph and a cat in a space suit as the input image. VAE Encode converts the image into latent data, and Repeat Latent Batch supplies 24 frames, enough for about three seconds at 8 fps.

The AnimateDiff loader connects to the checkpoint model, with Context Options Standard set to a context length of 16. The presenter recommends denoise between 0.6 and 0.8: lower values retain more of the original image, while higher values add motion but may change the character. The sampler is DPM++ 2M SDE with the Karras scheduler.

Video Combine receives the decoded images and exports MP4 at 8 fps. This local AI workflow takes longer as clip duration and frame rate increase, with generation time depending on the PC. The presenter suggests refining prompts and adding LoRAs to improve the result.