On a roll. Current thoughts on Mantella?

[anjenthedog]

Ok, one more for the mills.

 

What's the current "community" thought(s) on Mantella?  ("AI" NPC dialog)? It's an "interesting" idea, but it'd be nice to see what others who have taken the dive think... before I step out onto the spring board...

[traison]

Unless something groundbreaking has happened in the past year then until the following things are addressed I'm not becoming a user:

1. Delay. Having to wait 5-10 seconds is not acceptable. At the very least the generated fuz files need to be cached so that the wait is one-time only.
2. The use of xVASynth. Regardless of how impressive that application is, it still produces quite horrible voice lines unless you tweak every single character in a word. This could perhaps be sorted with extensive pronounciation files, or different software. Whatever the solution is, it can't require an online connection - I've had enough of phone-home software.

So for the moment I'm not switching to AI voice, regardless of how much I'd like to have it for my own mods with dialogue right now. I'd be more likely to pay for a day/week/month at elevenlabs and generate all voicelines I can think of in one batch. The reason why I haven't done that is that I don't really trust these AI services - wouldn't be surprised to find after paying the entry fee that every voice model costs extra. Sunk cost fallacy.

Edited May 5, 2025 by traison

[Vader666]

1 hour ago, anjenthedog said:

but it'd be nice to see what others who have taken the dive think...

 

It has the capabillity to turn skyrim actors into actual characters, adding a lot of dynamics to them and possibly the game itself.

However, LLMs are quite unpredictable, so your mileage may vary, depending on quite a lot of factors.

 

28 minutes ago, traison said:

Delay. Having to wait 5-10 seconds is not acceptable.

 

Response delay depends on quite a lot of factors.

Running the LLM and TTS localy, i get total response times between 3 and 10 seconds, using text input only.

 

34 minutes ago, traison said:

The use of xVASynth.

 

There are multiple options as TTS for Mantella.

In theory you can use any TTS you want.

 

[anjenthedog]

1 hour ago, traison said:

Delay. Having to wait 5-10 seconds is not acceptable. At the very least the generated fuz files need to be cached so that the wait is one-time only.

Good point I hadn't considered. So in principle it's pretty cool, but in practical implementation, it breaks down to "We don't have 400 GHz computers yet"

 

Also, yeah XVaSynth... not the worst of the two (lag vs robovoice), but still. my game is laggy enough already tbh, I don't want more. Plus robovoice

 

 

 

 

Edited May 5, 2025 by anjenthedog

[Parky]

5-10 second time lag is unacceptable, sheesh that's as real as it gets as you or your wife throttle the inside voice and try to say something constructive instead...........

 

All kidding aside. I wanted to ask the question of how do the various AI's react to Loverslab content ingame? Does it present a 21st century perspective on sex, domination, slavery etc or does it tailor it's responses to the fantasy world or to say a 14th century perspective?