Artificial intelligence generates videos, trailers and now commercials. Here's what's new

Enter a prompt and you get a video of a few seconds, which you can extend with subsequent prompts. But you can also: turn an image into a video, edit it with prompts by removing or changing background, people, synchronise the lip-sync in a video with new audio, etc. This is the list of functions that can be found - even for free - in the sum of artificial intelligence services specialising in video. Some only have part of these functions, which in any case tend to work better in some services and worse in others. We are at the beginning, although it is a fast-moving world.

This revolution started at Runway with the release of Gen-2 in February 2023, the first commercially available AI video generator. Soon after, Pika Labs launched Pika 1.0. The real breakthrough came earlier this year when OpenAI introduced Sora. Realism and motion quality took a big leap forward, thanks to the use of more data, more computing power and better algorithms. We have thus already entered the second generation of models, which you can try out for free, now with Dream Machine from Luma Labs. Sora is not yet available to the public in fact. Runway launched the second generation (Gen-3) in mid-June, but still only for paying users. As a result, the many videos we now find on social media, created with AI, are made with Dream Machine.

With this service, each generated video lasts about five seconds, but we can make other prompts to extend it by another five seconds. The free plan offers 30 generations per month, while paid plans start at $30 per month for an additional 120 creations. Note the long waits (several hours in our tests). The end result is very good, compared to the market average, especially in the handling of movement - the main problem of the first model generation. Of course, there are still distortions and anomalies.

Less good in this respect are the videos of Pika (first generation, still), which is, however, cheaper. The free plan provides 300 credits in total, while the one costs USD 1050 credits per month. Its strong point is the transformation of images generated by other services such as Midjourney or Ideogram into videos. It is also possible to use text prompts alongside image prompts. Pika Labs is working on a new, second-generation synthetic video generation model.

Runway's strong point, however, is its wealth of functions. In addition to classic text to video, there is video to video (changing the style of a video via a text prompt or an image) and various automatic editing functions. Highlights are the Motion Brush function, which allows only specific parts of an image to be animated, and a very effective lip-sync system, which can also animate head movements. Each video lasts about 4 seconds and can be extended up to 16 seconds. The free plan offers 125 fixed credits, while the basic plan costs $15 per month for 625 credits renewed every month. The second generation, as mentioned, is only available to those who pay.