A dynamic scene of a female martial artist in a white and pink traditional hanfu performing with a sword in a cobblestone courtyard. Onlookers in black martial robes stand in formation as cherry blossoms frame the ancient temple and tiled roofs, conveying a historic East Asian aesthetic and action-packed mood.

Woman in white and pink traditional hanfu wielding a sword amid… — AI video by @user_google_961350

@user_google_961350 · 5/17/2026

Another one of my samurai renderings that I felt has more dance in it than fight--and @CaylaCatz agreed. The original pic: https://cdn.budgetpixel.com/posts/14912/thumb/1774337745463807555-73f9ca6d-4bce-45ac-85da-162130c8a2d5.jpg

A dynamic scene of a female martial artist in a white and pink traditional hanfu performing with a sword in a cobblestone courtyard. Onlookers in black martial robes stand in formation as cherry blossoms frame the ancient temple and tiled roofs, conveying a historic East Asian aesthetic and action-packed mood.

Tags: traditional Chinese clothing, hanfu sword, martial arts performance, cherry blossoms, historical courtyard, ancient temple, martial arts drama, sword fighting, East Asian aesthetics, cosplay martial arts, wuxia scene, Chinese architecture, ceremonial weapon, period drama, cultural heritage

8 likes · 27 comments

Comments

CaylaCatz

I think there's some kind of interplay. The luck part is there is a seed which is a starting picture. The earlier models let you see the seed number so you could actually see what the starting image by putting nothing in the prompt. Now they hide the seed number in the later prompts. On top of the seed is us telling the AI what to do. If you give it a lot of instructions, it will minus everything that doesn't fit those instructions . If you give it less instructions, it will fill in the blanks by "guessing". I like the Ai doing some guessing because I am more likely to get something unexpected.

CaylaCatz

I was using SeeDance 2 and you have to put the music prompt inside the regular prompt. Copilot put it lower down but I put it up to the top because the higher something is, supposedly in some models, the higher priority. My first line was "Start from the provided image and create a samba dance video with epic samba music. " That's how I got the music in. But I wasn't trying to change sound midway or fuse different sounds together. It was a very simple sound prompt.

user_google_961350

There is a lot more control over the visual aspects of these than over the audio portion. If you specify color and material for something, for example, AI is usually pretty good at it. In contrast, I have tried to specify the genre of music or the kind of solo instrument I would like for the background music, and more often than not AI just ignores it.

CaylaCatz

I like to start with a start picture and maybe an end picture too so there is some control a little. but yeah for vid, I'm not good at prompting for sound. I'd rather do Suno and add the sound in but it's not worth it to do for 5 seconds

user_google_961350

In the prompt I asked for the music to be morphed into Irish when the guys joined in, but I guess that was too much for AI to handle. Anyway, yeah, I think this works just fine, too.

CaylaCatz

yeah, I'm not gooda at the sound aspect. But my last video I asked copilot how to add sound. So I did that way and the samba music came out pretty for the sama dance

user_google_961350

Yep--and it hasn't been that long. I think the first time I tried video generating with sound was when SeeDance put out their version 1.5, and that was about six months ago.

user_google_961350

So in the end, whether we give AI detailed instructions or tell it to use its own creativity and commo sense, it'll always boil down to luck. :-)

CaylaCatz

yeah, i noticed with very long prompts, sometimes the ai doesn't do it all but rather picks and chooses what sections/sentences it wants to do

user_google_961350

Maybe that's how the fight will start. 😊 "Hey, cut it out, guys. That's not the right dance." "Oh yeah? Says who?"