What type of model is appropriate for a plain stack of layers where each layer has one input and one output tensor?

The appropriate choice for a plain stack of layers where each layer has one input and one output tensor is the sequential model. This model is designed to allow you to build a neural network layer by layer in a straightforward manner, where each layer's output is the next layer's input. This structure enables a linear stack of layers to easily be implemented without any complexity in terms of branching or sharing of layers.

A sequential model is especially useful for problems where the architecture does not require multiple inputs or outputs and where the flow of data is in a single direction from the input through to the output. It's particularly beneficial for simple feedforward networks, where each layer is dependent solely on the previous layer.

In contrast, the functional model supports more complex architectures that may include multiple inputs or outputs, allowing for shared layers or complex branching paths, which is not suitable for a basic stack of layers. The convolutional and recurrent models each represent specific network architectures tailored for spatial hierarchies in images and sequential data, respectively, making them unnecessary for a straightforward stack of layers.