CONFORM: Contrast is All You Need For High-Fidelity Text-to-Image Diffusion Models