This paper presents RGB↔X, a unified diffusion framework for image decomposition into intrinsic channels (RGB→X) and synthesis from such channels (X→RGB). The intrinsic channels X include albedo, normal vectors, roughness, metallicity, and diffuse irradiance (lighting). The RGB→X model estimates these channels from RGB images, while the X→RGB model synthesizes realistic images from the intrinsic channels. The framework allows for flexible input conditions, including text prompts, and can handle heterogeneous datasets with varying available channels. The RGB→X model improves upon previous work by using more training data and adding support for lighting estimation. The X→RGB model synthesizes realistic images from intrinsic channels, supporting partial information and optional text prompts. The framework enables realistic image analysis and synthesis, demonstrated in the domain of realistic indoor scenes. The work contributes to the development of unified frameworks for both image decomposition and synthesis, with potential applications in material editing, relighting, and realistic rendering from simple scene definitions.This paper presents RGB↔X, a unified diffusion framework for image decomposition into intrinsic channels (RGB→X) and synthesis from such channels (X→RGB). The intrinsic channels X include albedo, normal vectors, roughness, metallicity, and diffuse irradiance (lighting). The RGB→X model estimates these channels from RGB images, while the X→RGB model synthesizes realistic images from the intrinsic channels. The framework allows for flexible input conditions, including text prompts, and can handle heterogeneous datasets with varying available channels. The RGB→X model improves upon previous work by using more training data and adding support for lighting estimation. The X→RGB model synthesizes realistic images from intrinsic channels, supporting partial information and optional text prompts. The framework enables realistic image analysis and synthesis, demonstrated in the domain of realistic indoor scenes. The work contributes to the development of unified frameworks for both image decomposition and synthesis, with potential applications in material editing, relighting, and realistic rendering from simple scene definitions.