My results: Video & sound: labeled all correctly as fake/real. Text: did not even try to read. My comments: For video, the biggest tell for me is the mouth.

My results: Video & sound: labeled all correctly as fake/real. Text: did not even try to read. My comments: For video, the biggest tell for me is the mouth. Just ~5 seconds of footage was enough usually. The lips shift in size and appearance or move way too little, for example. For audio, also a few seconds was usually enough to make a decision. The intonation of both Biden and Trump is way off in the fakes if you've spent some time listening to them during debates, interviews, etc. It has hints of the real voice but to me the fakes sounded quite a bit off. The curious fact is that I haven't listened to anything said by Trump since he left office but his characteristic voice is still very clearly embedded in my head. I guess humans are quite good at memorizing and recognizing voices. Biden to me sounds drunk/high in the fakes LOL. Overall, all of the fakes felt pretty easy to spot. Anybody with knowledge in the field: is this the current "state of the art" in terms of deepfakes or better results can be achieved?