I’ve also been trying this and would love a solid answer. I will say that ChatGPT (guiding Dall-e) and Sora both can generate remarkable, perfect sprites of any character in any pose… when they decide to. The issue is getting them to decide they feel like giving you the angle and pose you’re trying to describe for that frame of animation. It often takes repetitive queries, rephrasing, and dumb luck. And then your going to need to do some pallette matching and cut-n-paste in a graphic editor program. You absolutely can’t rely on them to create an entire sheet of multiple poses and get even a third of them correct.
Eventually I tried to get resourceful. I was hoping I could use Sora’s animation feature to just animate a sprite walking or rotating and then use the animation to mine still-frames, but this proved even more tedious than the still-frame process, as video just introduces an entirely new dimension of unrequested mischief the AI can pull on you.
If there’s a reliable, task-specific tool that takes a single frame of animation from a sprite and outputs an asymmetry-sensitive loop of idle, left foot, right foot in four directions, I would love to know about it. Or better yet, 8 directions, dead pose, attack wind up, and attack finish.