How to configure CNN and RNN?

[duvictor]

I'm trying to do a cnn+rnn architecture using autokeras, but i don't get success with this work. So, if someone can help me?
The input is a dataset with some images like plates with text. The images have 50x200px and the output has random text.

Follow some tries:
1:
input_node = ak.ImageInput()
output_node = ak.Normalization()(input_node)
output_node = ak.ConvBlock(kernel_size=3, filters=32, max_pooling=2)(output_node)
output_node = ak.ConvBlock(kernel_size=3, filters=64, max_pooling=2)(output_node)
output_node = ak.Flatten()(output_node)
output_node = ak.RNNBlock(layer_type="lstm", return_sequences=True, bidirectional=True)(output_node)
output_node = ak.RNNBlock(layer_type="lstm", return_sequences=True, bidirectional=True)(output_node)
output_node = ak.DenseBlock()(output_node)
output_node = ak.ClassificationHead()(output_node)
clf = ak.AutoModel(inputs=input_node,outputs=output_node,max_trials=5)

The follow error is showed:
ValueError: Expect the input tensor of RNNBlock to have dimensions of [batch_size, time_steps, vec_len], but got (None, 192)

I understand the output of the Cnn blocks is different from the inputs to the rnn, but i don't see any reshaping to fix the shape, so how can i solve this issue?