First test of Runway Gen-3

My first test of the new AI video engine Runway Gen-3 Alpha:
* Only text-to-image is available at the moment.
* The image quality is very impressive in many places and sets new standards.
* In detail, Gen-3 still produces a lot of errors and inconsistencies.
* I received more bad clips than good ones (which makes the high price twice as painful).
* The prompt conversion is very poor, at least for me. I hardly ever got the scene I described. More often than not, the result had hardly anything to do with my specifications.
* The realization of the described camera movement, on the other hand, is often accurate.
* Non-real fantasy motifs don't work at all.
Personally, I'm not really convinced by Gen-3 so far, but I see great potential and I'm very curious to see how Gen-3 compares to Luma.ai when an image-to-video is hopefully offered soon.
PS: I produced the soundtrack with Suno.ai, the original sounds in the individual clips with the video-to-sound function from ElevenLabs.

[@tierfilmer]

Without the promts, the clips are not very meaningful. Are these just the good results? Did you pay for a year right away?

[@adrianrohnfelder]

You are right but it was just a short test. I personally don't like text to video that much, prefer image to video. These are just the good results, yes. I would say 1 good result = 4 bad ones. And I only paid for one month and then bought additional credits one time. Still have some left as I am not really convinced.