FreeControl is a controllable method for text-to-image generation that can be used without any training. It supports simultaneous control over multiple conditions, architectures, and checkpoints. FreeControl aligns the structure of guided images through structural guidance and achieves visual similarity among generated images with the same seeds through visual guidance. The FreeControl system includes two stages: the analysis phase, where it queries the text-to-image model to generate a small number of seed images and constructs a linear feature subspace from the generated images, and the synthesis phase, where it applies guidance within the subspace for structural alignment of guided images and for visual alignment between images generated with and without guidance.