just figured out for 3 stage workflow you can reduce last stage to 2 steps each, change sigmas to 0.85, 0.5735, 0.0. Makes little to no reduction in quality from what I've seen so far with 10 second image to video, didn't try text to video yet. Distill lora was left at .6, maybe would need to be increased if there is a lot of movement in video. May as well leave the 2nd stage to 3 steps, unless it takes you a long time on your hardware.

Did you notice manual sigma is better than a static sigma?

Немного не понимаю смысл кучи апскейлеров, это если VRAM не хватает чтобы получить разрешение побольше?