>>100768608>They have no idea that I am a vtuberkek
>No way around for low VRAMIt's doable, you can use quantize versions and there are ways to offload if you have at least 32+ regular ram, generally gen speed decreases a lot. Though with Causevid LoRA (which lets you lower to 4 or 5 steps for a gen) and lowering the res a bit you could get around 8-10 min per gen, I think.
>optimizationHonestly hard to know but I think we maybe reaching a new minimum vram for stuff like this.
Anyway goodnight TWL