From Text to Pose to Image: Improving Diffusion Model Control and Quality