I'm trying to see how well Wan sticks to my prompts, so I'm doing multiple runs with the same settings/seeds but changing the prompts.

I tried with this image of Hu Tao that I found.

If I prompt "She looks to her left" she turns frame left, which is her right. Vice versa with "She looks to her right"

If I prompt "Another girl enters the video from the right." then nothing happens and I get a video of Hu Tao just kinda sitting there and moving slowly. If I add "She sits down on the bed" then another girl enters from frame right and sits down. If I change right to left then she enters from the left and sits down.

Now here's the part I wanted to test. If I say a character does an action with one of her hands, how does Wan understand it? For example, If I say "She makes a peace sign with her left/right hand" does that mean "her left/right" or our "left/right."

Well what I ended up getting was a bunch of videos of Hu Tao making a peace sign with her right (our left) hand, regardless of what I wrote in the prompt. Finally I got a video of her making a peace sign with her left (our right) hand if instead of specifying "left/right hand" I wrote in the prompt: "The anime style girl makes a peace sign with her hand that is on the right side of the frame. Her hand on the right side of the frame is in front of her making a peace sign. Her hand on the left side of the frame is on the bed."

So my main takeaways are:
1. Prompt adherence seems to improve with more details not in the sense that the movement gets more detailed, but that adding more details produces the movement at all as is the case for the prompt about a girl entering the video.
2. Directions (as in left/right) is ambiguous and may need more detailed prompting. Using frame based directions seems to produce the least ambiguity. No amount of prompting without frame relative directions produced a video of her using her left (frame right) without having to change the details of the movement in the prompt.