The model is failing because of a TensorFlow/Keras version mismatch, even if Python is the same, the libraries don't agree on how the weights were saved. The quick, reliable fix is to pin the exact version of TensorFlow/Keras you used locally like tensorflow==2.15.0 inside your requirements.txt file, then push to Render again.