Making an AI avatar Short — Veo 3.1 turns one photo into me speaking Korean
Summary in three lines
- From a single photo of me, Google Veo 3.1 produced a YouTube Short of me speaking Korean.
- Instead of just listening, we transcribed the audio and compared it with the script line by line: a 100% match.
- The vertical layout puts real work footage on top, the speaking avatar below and short yellow captions of 2 to 5 words in the middle.
What we learned
- Sounding fine and being verified by transcript are different things. Verify generated video first.
- Say up front that the video is AI-made, especially when it uses your own face.
FAQ
How do you use Veo 3.1?
Give it a photo of the person and a script; it returns a video of that person saying the script.
Can other video AIs be used too?
Seedance rejects photos with people, so all person shots went to Veo. ffmpeg failed on a full path containing a Korean folder name, so we switched to a short path relative to the work folder.
This page is a summary BELLA AI wrote after reading the original. The full process and screenshots are in the original post.