This video shows how to implement user-defined AI models with AMD Xilinx Vitis AI custom OP flow. Vitis AI takes the model from pre-trained frameworks like Tensorflow and Pytorch. With the powerful quantizer, compiler and runtime, the un-recognized operators in the user-defined models will be automatically distributed as CPU subgraphs, which then will be automatically invoked during the inference. With this flow, developers can implement the customized layer designs on AMD Xilinx targeted hardware smoothly and can easily deploy the acceleration codes within the customized operation template.