X-Adapter is a universal upgrade tool that allows pre-trained plugin modules (such as ControlNet, LoRA) to be directly used with upgraded text-to-image diffusion models (such as SD-XL) without further retraining. By training an additional network to control the frozen upgraded model, X-Adapter retains the connections of the old model and adds trainable mapping layers to connect the decoders of different version models for feature remapping. The remapped features serve as guidance for the upgraded model. To enhance X-Adapter's guiding ability, we employ an empty text training strategy. After training, we also introduce a two-stage denoising strategy to adjust the initial latent variables of X-Adapter and the upgraded model. X-Adapter demonstrates universal compatibility with various plugins, enabling different versions of plugins to work together, thus expanding the functionality of the diffusion community. Our extensive experiments demonstrate that X-Adapter may have broader applications in upgrading basic diffusion models."