/vt/ - Virtual Youtubers

>>84869770
I'm sorry I didn't mean it like that, I meant if someone wanted to get the original prompt from a specific timestamp on the video I posted, but that's my fault for wording it wrong.
The hard part is mining good pictures and being patient with slow generations.
I suggest going on kemono.su and searching for the characters / artists you like and include "koikatsu" because they tend to shit out a lot of pictures in groups and I personally believe 3d anime images works better than 2d anime images (but I haven't really tried using 2d anime much). Kling fucks up eyes, Luma tends to fail at latching onto the character if it's slightly complex, when it works, it's usually not hot and vidu and runway will probably censor the image (but when runway works, it's pretty nice, but my prompts are boring, I would play around more but fear of wasting tokens and I can't queue so you need to retry periodically).
If i wasn't so lazy I would probably use koikatsu myself (i'm sure it's not that hard to pirate), but at that point you could forget about AI video and probably just make a looping animation using koikatsu, because it takes some time to find a picture, retry like 5 times, give up on the image, repeat. AI video is not practical at all, it's not worth paying, and the free tiers are getting shitter, they block sexual stuff, etc.
This is an old prompt, it also helps if you try to add words to define the position the characters if you have problems with mismatching, and you could probably get away with writing less text, also note other models hate long descriptions, that's just kling. Also I have not done enough A/B testing, I personally don't care about figuring out what makes a good prompt because the next AI video model will work better with completely different rules.