CogVideo for experimenting with video diffusion models

How it works

The repository describes checkpoint families and their supported inference paths. Text and image conditioning options depend on the model you select.

What you need

Choose the CogVideoX checkpoint for the input mode you want to test. Set resolution and duration according to that variant rather than borrowing settings from a different example.

What to try first

Generate two short clips with the same subject and different camera instructions. Compare temporal consistency and how well each follows the prompt.

Compare the text-conditioned and image-conditioned workflows only with matching supported variants. They can have different model and preprocessing requirements.

Where to look for failures

Different checkpoint variants need compatible pipelines. Check input dimensions, acceleration options and distribution terms before building a larger workflow.

Related Franklin pages

Compare complete clips through playback and frame inspection. Keep prompt, duration and dimensions as similar as supported, and inspect whether subject and camera motion match the task.

• Experiment with the official Wan text-to-video and image-to-video workflows. Wan 2.1 for running video generation checkpoints
• Run documented generation pipelines for text and visual conditioning. LTX-Video for prompt and image-conditioned clips

For video assembled from authored HTML and motion graphics, see Franklin’s HyperFrames profile. HyperFrames

Project documentation

• Current project README
• README at the reviewed revision