How to add bias variable in Keras with functional api without subclassing?

Question

I tried the following approaches:

import tensorflow as tf
inp = tf.keras.Input(shape=(None,))

# first approach:
bias = tf.Variable(0., shape=(), trainable=True)