CogVideo ImageEncode

This is the node that makes image-to-video happen. You give it a starting image, it encodes that image through the VAE into a latent, and that latent becomes the anchor the sampler animates forward from. Instead of dreaming a video out of pure text, the model starts from your picture and moves it. For a lot of people that's the whole appeal of video generation - I2V is the community favorite precisely because anchoring the first frame to a known image sidesteps most of the consistency headaches that plague text-only generation.

You need an I2V-capable model for this to mean anything - one of the -I2V variants from the loader. Feed a plain text-to-video model an encoded image and it won't know what to do with it.

The inputs that matter

• vae - the VAE from your loader. Required, because encoding an image to latent space is literally the VAE's job.
• start_image - your first frame. This is the picture the video grows out of.
• end_image (optional) - give it a second image and you get first-frame/last-frame behavior: the model interpolates a motion path from your start image to your end image. Great for controlled transitions.
• noise_aug_strength (optional, default 0) - adds a touch of noise to the conditioning. Small amounts can loosen the model up and give you more motion; too much and it drifts away from your image. Start at 0 and nudge up only if the result is too static.
• strength (default 1), start_percent (0), end_percent (1) - how strongly and over what part of the schedule the image conditioning applies. Defaults are the standard I2V behavior.
• enable_tiling (optional) - tiles the encode to save VRAM on large images.

The output, samples (a LATENT), wires into the CogVideo Sampler's image-conditioning input (image_cond_latents) so the generation starts from your frame.

How to install it

Comes with the wrapper:

• ComfyUI Manager - search ComfyUI-CogVideoXWrapper, install, restart.
• Manual - cd ComfyUI/custom_nodes && git clone https://github.com/kijai/ComfyUI-CogVideoXWrapper, then pip install -r ComfyUI-CogVideoXWrapper/requirements.txt, restart.

Common issues & troubleshooting

The video barely moves. CogVideoX I2V can be conservative, staying close to the input frame. Nudge noise_aug_strength up a little to inject motion, and lean on your prompt to describe the action - motion comes from the prompt as much as the image. Describe what happens, not just what's in the picture.

The output drifts away from my image. The opposite problem - usually noise_aug_strength too high, or strength too low. Bring the augmentation back toward 0 and keep strength near 1 to stay faithful to the start frame.

It errors or ignores the image entirely. You're on a text-to-video model. This node only does anything with an -I2V model, and the encoded samples have to be connected to the sampler's image_cond_latents slot. Both halves matter.