This simple workflow allows you to generate Videos with Wan in a much faster way with CausVid V2.
All you need to do is put the Causvid v2 lora (already included) at the end of the loras selector and you can now generate in only a few steps.
Image Loader and resize :
Input an image in the loader, select the dimensions (Advised to use 480x832 for 480p and 720x1280 for 720p models).
Models Loader nodes :
You can choose different model if you prefer, the Unet Loader load the GGUF wan model (Here the 480p Q8 model is used), the clip loader uses fp16 but you can use the fp8 for faster generation but at the cost of quality. the VAE and Clip vision don't need to be changed.
Lora & Prompt Nodes :
The first node allows you to load multiple loras, but the one you need is the CausVid T2V (also i2v) V2 already put in it. that Lora seems best at strength 1 but you can try different weight if it needed, but a lower weight seems to kill the result's motion.
The Prompt Stash Saver is only an utility in case you want to save certain prompt for certain Loras.
Text encode : Just put the prompt, but do not remove the chinese negative prompt, it's pretty much vital for wan.
Generating Nodes :
The main node you will edit is WanImagetoVideo for the length of the video, 16fps so 81 = 5 sec.
You can also change the steps count, 8 is a good compromise for speed at a pretty good quality, but feel free to experiment. The first Ksampler allows to drive the video mainly to avoid the motion killing of CausVid by using a normal cfg, at 8 steps, it is set to end at 3 steps, which is then continued by the 2nd sampler on the right, which starts at 3 steps to finish the video with a cfg of 1.
You can try different sampler and scheduler (use wan compatible ones tho !) but you will probably stick with these ones. you can try different cfg weight for the first sampler but the second one should stay at 1. and you can also try different combinations, of steps as long as the samplers start and end at corresponding steps.
Output Nodes :
The video finally get combined into 16fps in the first node and then go through FILM VFI to be interpolated at 32fps.
The Folders are already set so you can put your models and loras in them.