We have some VIPs coming to visit Duke’s Innovation Co-Lab and while at a meeting this morning, I noticed there wasn’t some really compelling content on the digital signs. I found out the visit was tomorrow, so I asked the head of the lab to send me some photos.
Since these “actors” were already cleared for publicity photos and the video would only play inside of the Co-Lab, I deemed it OK for me to animate them. They are being published via YouTube as unlisted and are being broadcast as a demonstration and education.
I used Google Flow with a very simple prompt, “Animate”.
I didn’t see a way to get rid of dialog but the first test, the dialog mouth movements were pretty bad:
I added to the prompt, “Animate. No dialog. Minimal Movement” – some of the movements were exaggerated and that prompted helped it.
Here’s the draft of the video.
The video quality was extraordinary. They are 1200×800 images so I didn’t have time to use Photoshop extend to make it 1920×1080 so I added the background on the sides.
The logo was interesting too. I loaded the logo with the white background and just told it to change the background into random colors and patterns and it did a GREAT job.
There is music as well. I used Google’s Lyrica 3 within Google Gemini. It did OK. Not nearly as controllable as Suno but it fit the bill of doing something quick. I used about 3 prompts to get the EDM track that I wanted. I think it was more about prompting and not as much about reaction but the tool kept saying, “I’m still practicing my music editing skills, …” Here’s the session and the different options.
I ended up keeping with the one that is third in the video above. There were two additional audio files but they didn’t get better. I was trying to tell it to play a more definitive ending of the one I liked, and it could only create a completely new song. The first one did have a definitive ending, but was only 20 seconds. I tried to extend it.
90 minutes of work. Pretty good for this particular application.
Subjectively, it was thought that the likenesses of actual people were too good – that uncanny valley that people talk about. I thought I’d simply change the people and be done. Well, from the time of my first version to starting the people replacement part, Veo changed the interface completely and refused to work.
Ugh. I used Gemini (not Veo) to change the people, that worked well. You can’t create video in Gemini so I went to Seedance (owned in China by ByteDance – TikTok). It worked fine but the quality was not nearly as good as Gemini.
This was a fun blooper too. I think they got a little crazy when I tried to animate an image of the laser cutter:
It was an interesting process but due to the failure of Veo and the lower quality of Seedance, I scraped the project. Was a great test however!
