Image generation: Limb placement for different walking steps

I want to generate images for some platformer mini-game. My main problem: I am unable to generate two different step walking sprites. You can see the image below to understand the problem.

I know AIs are bad at limb placement but I assume that since, nowadays, AIs are able to generate videos, they should at least be able to generate walk_a and walkd_b images. Here is the list of things I tried:

1. Create everything in one go while explaining the high-level context (two walking images for different steps in a video game.
2. Try to be extremely explicit about limb placement (which leg and arm should be forward and backward or right/left).
3. Ask him to generate walk_a first and then walk_b, one by one.
4. Using an extremely rare and unique (by chance?) example where I managed to make it work in the style I want to ask him to do “pose transfer” or “pose matching”.

Additional note: I noticed that the image generation is taking EXTREMELY long. After looking at what the model is thinking and directly asking him, it is because while he understands his task, the image generation model keeps collapsing into only 1 walking frame and therefore the AI tries again and again and keeps failing.

I am kind of giving up and hope you guys can help. What is the best approach to make it work? I truly think it should be possible to make it work regularly. I am thinking that maybe I should use some stick figures to help with the limb placement…