Tried out Vidu, another AI img2vid site. It's pretty good.

As expected, it sucks at teeth. Water is wet.

Other than that, it strongly follows what you asked for, albeit at the cost of doing it a bit too literally. So, from my experience, I added in words like "mildly" or "slowly" since if it's just something like, "She is laughing.", the AI will literally lose itself and make the image as well lose itself.

By putting "She is mildly laughing.", it helped give me this much better result.

I also like how "two stepped" it feels. It's not like Hailuo or KLING where it's smooth and slow (not that it's a bad thing no lol). It feels more experimental and while that can lead to more obvious artifacts, I think what it's going for somewhat makes up for it.

Overall, it's epic.