Most of the models currently available for Neutone are timbre transfer models, which transforms an input sound timbre to another. Typically you’ll get better result with single instrument input (or voice) and with DDSP models, you should use monophonic (no chords) input.
We also saw better results when the input sound is filtered towards certain frequency range where the model is more sensitive to responding. It is also a good idea to use a limiter or compressor to saturate the dynamic range for better responses. Try for example using set of effects like the following: