>>19842606honestly i find the best way is to find a video sample on civitAI and hope it has a comfy workflow embedded, and drag it onto the comfy ui to load it, to see how it was generated. of course it doesn't help that there's a million variables affecting a gen, like the seed (obviously), model, loras, each lora strength, order of chaining loras, positive and negative prompts, WAN video resolution, num of frames/duration, no of steps, sampler, cfg, the list is fucking endless, at one point i literally just picked one set and I only ever change the prompts (and occasionally the loras).
>>19842811np bro :)