writes:It might surprise you to know that the most popular videos we make, by category, are the staff editor videos. Far and away the most viewers, here and on social media. Great for reaching out to the larger community.
Therefore, it made a lot of sense to make another video. Former Editor Ellah was eager to have another turn. But I was not completely happy with the technology we used to make the earlier videos, and wanted to try something new.
What I wanted to try was to combine the narration with actual examples of miniature wargaming. The topic of the new video, continuing in the vein of her previous video, was to explore boardgames and miniature wargames and their use of maps or tabletop terrain. We quickly worked out concepts and a script for a 30-second video.
My first thought was to use AI to have Ellah 'walk through' various wargames and miniature wargames, as she gave her narration. A full-figure image of Ellah was created, and we experimented with animating it against various backgrounds. Unfortunately, the AI tools I'm familiar with could not handle that (or I couldn't figure out how to make them do it!).
However, while exploring options, I discovered that ClipChamp – the Windows software I use to compile our videos – supports 'picture-in-picture'. That meant I could superimpose a 'talking head' on top of a slideshow-type video showing examples of games.
Making the Pieces
To make the simple video we had planned, I would need four background pictures, and three animations of Ellah speaking (to add up to 30 or so seconds of narration, since the tech we're using is limited to producing 10-second videos).
A new head-and-shoulders portrait of Ellah was made (based on the same original photo), and AI created three videos of her performing the narration.
For the example photos, my first thought was to simply ask the AI to generate the needed photos from scratch. This proved difficult, as AI simply did not understand enough about boardgames or miniature wargames to produce accurate photos.
Above, you can see the AI-generated example of a boardgame played with cardboard counters on a map. It was good enough for an example, but doesn't look like a real boardgame. Some of the counters are not within the grid. Why are there fortification pieces? The limitations of what AI knows…
I then resorted to using existing TMP gaming photos, and using AI to turn them into what was needed.
Example of a miniature wargame. The above photo is from an actual game, but with the participants replaced by the AI. (The photo was taken at a convention, in public, but I felt it better not to use the original people for a video without their knowledge. The AI generated two college-age gamers.) I also asked the AI to replace the visible cellphone with the latest model!
Example of a boardgame played with miniatures. Background modified and players inserted by AI. It took a lot of iterations to get it correct, and in the process it became more and more like a painting, and I could not get AI to turn it back into being photo-realistic.
The final example: a miniature wargame played on a grid, similar to a boardgame. This was an existing TMP photo, clipped and resized, with AI used to improve the quality after the photo was enlarged.
Making the Video
I first created the video slideshow of the four background pictures.
I then used ClipChamp's 'picture-in-picture' mode to put Ellah's narration in a box on the left side. I hadn't thought about how the narration box would work against the background photos (!), so tested to find the right position to put the narration box.
One minor problem came up when I decided not to use the default box size, but to enlarge the size. There was no tool to enlarge all three videos to the same size. Instead, I had to play 'hunt the pixel' to get the videos to line up.
Caption files were created for English, French, German, Spanish and Tagalog.
And then the final video was created.
I've posted the video as a reel on the TMP Facebook page. It has been one of our most popular videos ever, with much positive reaction. Surprisingly, 90% of viewers came from the TMP feed, with 78% being suggested by Facebook. It was most successful in the 25-34 age bracket, with top countries being United Kingdom (30.7%), USA (25.2%), Australia (5.1%), France (5.9%), Italy (4.4%) and Poland (4.1%). I don't remember seeing Italy or Poland in our results before, and surprised not to see Canada this time.