Consistent View Synthesis with Pose-Guided Diffusion Models