Generative Photography: Scene-Consistent Camera Control for Realistic Text-to-Image Synthesis

2 NVIDIA 4 scene while modifying only the camera settings to achieve varied photographic effects. Current state-of-the-art text-to-image generation models like Stable Diffusion 3 (SD3) [4] and FLUX [1] face two major limitations: failure to accurately interpret camera-specific settings and difficulties in maintaining consistency in the base scene. This paper introduces a novel approach that addresses these issues, enabling precise camera setting control and maintaining scene consistency in generative models.