@DocentHorsecock
After experimenting around on some other site, I came to the conclusion that the AI just doesn’t know what to do with her coat. It’s draped over her shoulder like a cape, but has details like a coat, or even a jacket. Tried all three of those words and it just wouldn’t do it. But calling it a cape gave the most consistent results. As for the sword, the first thing I tried is specify that she’s drawing it with her left hand, to make the motion more natural. But that seemed to make no difference. What did make a lot of difference is describing how she stands afterwards.

The prompts: “The person in the image takes her cape off, then she draws her sword and stands ready to fight, with a confident expression” gave decent results, but still not quite there.

@Background Pony #554C
This is possible…

@Draco Dei
Maybe it was trained on lightsabers, lol.

@JesterOfDestiny
Hmm I guess this encourages me to try to write longer prompts. Btw this is WAN 2.2, and there is WAN 2.7 already. No idea how much better it is. I know Grok does really good stuff, like that Sunny:

Haven’t tried grok myself. I guess this Zipp could be a good test. Maybe next time.

Give me your best prompt for this and I will try it on WAN 2.2 :D

Zipp didn’t just learned to phase her sword through the scabbard, but also duplicate it!
Pegasus magic!
(But it looks good for the rest, gj)

@Draco Dei
One thing I’m sure of, is that AI can be very particular about instructions. It’s just like coding; you have to tell the machine exactly what to do. Your vision might be that she throws her coat down, but “flings her coat aside” could also just mean she brushes it out of the way. Try to find the exact words that will make the machine understand what you want it to do. Be as detailed as you need to be. Perhaps “takes her coat off and throws it away with a quick dramatic motion” might work better. And also, what how does she draw the sword and what does she do once she’s drawn it. Is she pointing it at someone? Is she threatening? Does she make a big dramatic display, or is she focused?

It’s like being a director to an actor, who doesn’t understand what’s it like to be human. Of course, the machine itself has some inherent limitations and some randomness is to be expected. Maybe it gets it right after you hit generate the sixth time. And maybe it just hasn’t been trained in the specific thing you’re trying to do and it’s gonna take a bit more training for it to understand what unsheathing a sword looks like.

Nice to be (I assume) inspirational, but good grief does the AI not understand “unsheathing a sword” and “dramatic cape fling” isn’t much better. The fact the eye contact and sword pointing are in different directions is almost certainly a consequence of the original image not having the head pointed forward, and body rotation probably being “expensive” for generations. Could maybe try instructing it to rotate the HEAD, if you think it is worth the time and money/points.

Oh, hell yeah!