It seems to basically be using an image captioning model like this one: https://deepai.org/machine-learning-model/neuraltalk https://deepai.org/machine-learning

It seems to basically be using an image captioning model like this one: https://deepai.org/machine-learning-model/neuraltalk https://deepai.org/machine-learning-model/neuraltalk , but possibly trained on more data? I wonder what dataset they’ve used, or if they created it themselves and won’t share it.