This video was created using a 3 steps total workflow in 720p, around 25% faster than normal 2 steps per model workflow. The idea was that since the first high noise model is high noise anyways, there may be a configuration of parameters that only needs 1 step for it. It seems to work but I have mention that from time to time the image gets blurry and have not tested it other with images from THESE series witch have a very particular style and the motion is very still.

Here is the workflow, you can try it yourself. Note that you need the lighting loras and the Wan2.2 I2V Models (I am using GGUF versions). Any missing nodes should be downloaded using the manager.
https://aurelm.com/upload/ComfyWorkflows/Wan_22_IMG2VID_3_STEPS_TOTAL.json