This version was trained with the latest reprocessed dataset.

We also changed the trigger words to give the distinctive pictures a noisy feel.

Use your favorite of the two versions.