AI videos from still CGIs

Here are just a few of my recent experiments with creating videos using Kling AI, I have generally been using my own computer generated still images as a reference. This helps lead to consistency in the buildings between prompts, and importantly allows me accurate control over the actual building design. The technology is not perfect yet; still sometimes giving unpredictable results, and a few glitches, morphing and wobbly bits here and there. They probably don’t stand up to close scrutiny on a large screen. But that said, I think some of the results are pretty exciting, especially considering the relatively short time they take to create.

The short video shown above was created from this single CGI on the left. Prompted three separate times; to create the two camera movements, and then the time-lapse into evening effect.

The night time CGI to the right was used in conjunction with these reference images below…

…to create this video on the right. Despite the odd weird flickering and morphing in a few places, I was pleasantly surprised how it creatively combined these elements, while still keeping consistency in the architectural detail of the building.

The video “House” above and on the left, was created from four separate still CGIs (below) created of the same house. This one did take quite a lot of tries, with several rejected videos, for some reason it just kept adding random planting, where I didn’t want it. However with a little editing I was able to put together a fairly consistent looking short video.

Again, these four still CG images below were used to create the video below them, I wasn’t completely happy with the look of the traffic outside in the first shot; it has a very “ai” morphing feel to it, but overall the video quality isn’t too bad.

Going back in time a little bit with the video on the left, “MF 003”, these still images were my first ever freelance project. I always thought the rear views of the property weren’t quite as strong as the front views, but the ai video seems to have done them a favour; making them look quite convincing now.

A more recent project on the right…

For this video above I used a combination of different prompts using both the ‘Frames’ and ‘Elements’ starting points: ‘Frames’ results in videos that very accurately represent the reference images, which are great for things like buildings and furniture, where you need the objects within the video to look exactly how you visualised them in the CGIs. However to introduce some more dynamic lifestyle-type shots, with people in, then I’ve used the ‘Elements’ feature: This introduces a bit more risk, and plenty of parts with strange movements and things added you didn’t really want, which need to be edited out. This video is far from perfect, but with more time, and an unlimited number of prompts you could probably get something with less errors in it.