**Don't forget to Like 👍 the model. This inflates my ego and costs you nothing! Haha ;)

*Known bugs from CogVideoX (At least for me...):

• Allocation on device: If you're having this error most of the time you can solve it by using a fixed seed and just Queue-ing the workflow again without changing anything. It will continue from where it stoped and will return the result. This error usually happens in CogVideo Decode node.

• Taking too long to run each Sampler: While running the Samplers, if it's running at a value much greater than 10.00s/it (eg: 200.00s/it), something went wrong while loading (I don't know what... Just some weird bug). Stop the process and reload your Comfy.

This workflow allows you to create image-based videos. Besides the basics, it also lets you extend the video duration up to seven times (although, to be honest, the result isn't great most of the time). And best of all, it works with little VRAM!

I originally created this workflow to be used with SVD models, hoping that at some point the model would evolve enough to prevent the quality of generated frames from deteriorating, so that the video could continue. With the CogVideoX model, the deterioration is much less, allowing the video to be continued from the generated frames. So, I adapted it for this model.

Unfortunately, the dimensions of the generated videos are small regardless of the configuration. I considered adding a method to improve the final result, but I don’t want to overcomplicate things. Overall, the quality of objects in the videos is good. People, on the other hand, deteriorate a lot, and the movements look unrealistic. Personally, I really like using this model to generate landscapes. It works very well! That’s it. I hope you like it. Don’t forget to leave a Like on the model!