thanks mate! Consistency is indeed a problem. The only way in my head to solve it atm is to include similar style image input if you are doing image-to-image, image-to-video gen.