UPDATE:

I've been getting better results with the lightx2v lora distill from kijai (https://huggingface.co/Kijai/WanVideo_comfy/blob/main/Wan21_T2V_14B_lightx2v_cfg_step_distill_lora_rank32.safetensors), which makes the 2 sampler setup obsolete because this works great at CFG 1, so this is just a quick update if anyone wants to use lightx2v with the same workflow layout and settings that work for me.

ORIGINAL:

I've seen people talking about using 2 samplers to increase the movement/quality of CausVid, but no one's posted a workflow yet so I thought I'd share, because I've been impressed with the results.

The basic idea is using the KSampler Advanced which lets you split the generation into 2 or more samplers with a start and end step, you generate the first 3-6 steps using a lower CausVid strength and high CFG, then send the latent with leftover noise to a 2nd sampler to do the rest of the steps at a higher CausVid strength and low CFG.

I've kept the workflow simple, it's adapted from the workflow I typically use for my Wan videos. Should work fine with any of the 720p/480p I2V models, adjust the sliders as necessary.

I'm not sure what the best combination of sampler/CFG/steps/CausVid strength is yet but this one seems to work well, let me know if you find any improvements.