Slowly figuring out optimal settings for Wan2.2. Probably need more steps or a more aggressive sampler/schedule idk. This is 20 steps with Euler/simple, and the faster-moving bits still have significant motion artifacts. The coloring trends toward more saturation, which may be a prompting issue. The upscaler change from 4xR-ESRGAN-anime over to 4xUltrasharp reduced a lot of temporal smudging in the details. First impressions are still pretty good though; fewer incoherent gens and better prompt adhesion, plus it still takes Wan2.1 LoRAs.