MinAI - AI (SFW/NSFW) for CHIM

[Loopdeloop]

Damn, this moved fast!

[Loopdeloop]

Since installing the mod I've been getting huge spikes in script latency, jumping to over 600 ms. This happens every 30 seconds or so, and causes the game to freeze for around 20-30 seconds. 

[Min]

25 minutes ago, Loopdeloop said:

Since installing the mod I've been getting huge spikes in script latency, jumping to over 600 ms. This happens every 30 seconds or so, and causes the game to freeze for around 20-30 seconds. 

Which version? The AIFF version? Share your papyrus log?

[Loopdeloop]

5 minutes ago, Min said:

Which version? The AIFF version? Share your papyrus log?

Decided to test it on a new playthrough rather than an existing one. So far it seems fine, will let you know if I have any further problems. 

[Min]

53 minutes ago, Loopdeloop said:

Decided to test it on a new playthrough rather than an existing one. So far it seems fine, will let you know if I have any further problems. 

Ah, yeah. Beta14 was an enormous update - Due to renaming the esp, it's probably not savegame compatible without work. 

[crazyduck]

I woudl not call AI Follower Framework superior really not. It does not support xvasynth so you need to pay for text to speech or use mimic with horrible voices or let xtts run from the virtual machine and wonder why nothing happens cause it simply throw away a response if processing takes longer as 30 seconds.

And if i compare both, with mantella i have much more interaction with my followers as i have with AIFF, with AIFF there is most of time silence , the followers to not banter they do not comment on anything and if i ask them something i get no repsone or the narrator starts to speak as i have asked him and not the follower.
Also that AIFF does not has a working console like mantella so i coudl see if something is wrong somewhere is a bad thing, with mantella you see in console if somethign is wrong , with AIFF you only have the command window open with a fixed symbol and the startup information that is not really that much.

[Loopdeloop]

Papyrus.0.log

1 hour ago, Min said:

Ah, yeah. Beta14 was an enormous update - Due to renaming the esp, it's probably not savegame compatible without work. 

Same problem on a new character, unfortunately. Attached Papyrus log.  Resaver reports 

 

<a href="script://minai_AIFF">minai_AIFF</a> occurs the most often as a stack frame (222 occurrences)</p>

Frame count

<a href="script://minai_ContextEffect">minai_ContextEffect</a> occurs the second most often as a stack frame (218 occurrences)</p>

 

Edited September 15, 2024 by Loopdeloop

[Min]

44 minutes ago, crazyduck said:

I woudl not call AI Follower Framework superior really not. It does not support xvasynth so you need to pay for text to speech or use mimic with horrible voices or let xtts run from the virtual machine and wonder why nothing happens cause it simply throw away a response if processing takes longer as 30 seconds.

And if i compare both, with mantella i have much more interaction with my followers as i have with AIFF, with AIFF there is most of time silence , the followers to not banter they do not comment on anything and if i ask them something i get no repsone or the narrator starts to speak as i have asked him and not the follower.
Also that AIFF does not has a working console like mantella so i coudl see if something is wrong somewhere is a bad thing, with mantella you see in console if somethign is wrong , with AIFF you only have the command window open with a fixed symbol and the startup information that is not really that much.

 

It supports xtts, just like mantella does. My mod actually has an optional feature to make AIFF work with mantella's xtts setup.  xtts is generally better quality than xvasynth anyways. I'd recommend running that via runpod though (I'm using Mantella's xtts template), rather than locally.

 

Sounds like your setup is broken, or otherwise misconfigured. AIFF writes a lot more info to logs, making it easier to debug than mantella's console. It's fine if you disagree, but I think you're wrong.

 

[Min]

4 hours ago, Loopdeloop said:

Papyrus.0.log 253.32 kB · 1 download

Same problem on a new character, unfortunately. Attached Papyrus log.  Resaver reports 

 

<a href="script://minai_AIFF">minai_AIFF</a> occurs the most often as a stack frame (222 occurrences)</p>

Frame count

<a href="script://minai_ContextEffect">minai_ContextEffect</a> occurs the second most often as a stack frame (218 occurrences)</p>

 

 

I think the system to update context is putting too much load on your script engine in that version. I just pushed a small little release here: https://github.com/MinLL/MinAI/releases/tag/beta15 that improves performance a fair bit. See if that helps?

Beta15 has a serious bug - took it down for the moment. The rest of this still applies though.

 

You can also remove the MinAI_DISTR file that ships with the mod to disable the context effect, which is where most of the script load comes from. See if that helps, if beta15 doesn't.

 

Send me a copy of your papyrus.0.log on beta15 if that doesn't help. 

Edited September 15, 2024 by Min

[crazyduck]

5 minutes ago, Min said:

 

It supports xtts, just like mantella does. My mod actually has an optional feature to make AIFF work with mantella's xtts setup.  xtts is generally better quality than xvasynth anyways. I'd recommend running that via runpod though (I'm using Mantella's xtts template), rather than locally.

 

Sounds like your setup is broken, or otherwise misconfigured. AIFF writes a lot more info to logs, making it easier to debug than mantella's console. It's fine if you disagree, but I think you're wrong.

 

ok i use xvasynth with the voices and i get most of time a instant voiced answer, i tried with the xxts2 installed by AIFF and i got nothing most of time and i could see in the log that processing a longer ansewr took 35 seconds so it was simply ignored.   Also the LLM selection, that i can not see in AIFF which model i choose in the list and only show the prices for each model until i select one is a bad thing, its horrible to click 20 times on "get model list" select one of the many models to see "oups no that is the wrong one" mantell shows me the list with Model Name and pricing.-
I will give AIFF another try and will see if the last update of the mod makes it work better.

[Min]

3 minutes ago, crazyduck said:

ok i use xvasynth with the voices and i get most of time a instant voiced answer, i tried with the xxts2 installed by AIFF and i got nothing most of time and i could see in the log that processing a longer ansewr took 35 seconds so it was simply ignored.   Also the LLM selection, that i can not see in AIFF which model i choose in the list and only show the prices for each model until i select one is a bad thing, its horrible to click 20 times on "get model list" select one of the many models to see "oups no that is the wrong one" mantell shows me the list with Model Name and pricing.-
I will give AIFF another try and will see if the last update of the mod makes it work better.

 

Yeah, that's not how it works when it's set up correctly - I actually have sub-1 second response times from when I say something, to when the AI responds with the voiced comment depending on the model I'm using. Far faster than I ever managed to achieve through Mantella, despite a lot of messing with it.

 

The setup for AIFF *is* much more complicated than the setup for Mantella. This is the only real drawback to it. Once setup though, I can't think of a single thing that Mantella beats it at. It doesn't come close in most ways, either from a developer perspective, or a user perspective.

Edited September 15, 2024 by Min

[Min]

 

57 minutes ago, Loopdeloop said:

Papyrus.0.log 253.32 kB · 1 download

Same problem on a new character, unfortunately. Attached Papyrus log.  Resaver reports 

 

<a href="script://minai_AIFF">minai_AIFF</a> occurs the most often as a stack frame (222 occurrences)</p>

Frame count

<a href="script://minai_ContextEffect">minai_ContextEffect</a> occurs the second most often as a stack frame (218 occurrences)</p>

 

Also - FWIW, it may just be a matter of "This new release is more script intensive than the old one and it's pushing your papyrus setup over the limit" (Increased script load is mostly due to that script for npc context awareness I mentioned). I do see from the log that you're running a couple other really script heavy mods - Namely, "Wet and Cold" and XPMSE's styles script. That XPMSE script in particular is notoriously poorly performant. You don't need to run the XPMSE esp / scripts to use the XPMSE skeleton that mods require.

[crazyduck]

1 hour ago, Min said:

 

Yeah, that's not how it works when it's set up correctly - I actually have sub-1 second response times from when I say something, to when the AI responds with the voiced comment depending on the model I'm using. Far faster than I ever managed to achieve through Mantella, despite a lot of messing with it.

 

The setup for AIFF *is* much more complicated than the setup for Mantella. This is the only real drawback to it. Once setup though, I can't think of a single thing that Mantella beats it at. It doesn't come close in most ways, either from a developer perspective, or a user perspective.

Ok AIFF works now , you said somethign about a patch to make AIFF use mantella XTTS i have the mantella XTTS up and running now but AIFF webinterface tells me the url is correct but it cant connect to it.

[Min]

13 minutes ago, crazyduck said:

Ok AIFF works now , you said somethign about a patch to make AIFF use mantella XTTS i have the mantella XTTS up and running now but AIFF webinterface tells me the url is correct but it cant connect to it.

 

The URL for that should be https://<your-pod>.proxy.runpod.net

 

Follow the mantella setup instructions for getting a mantella xtts runpod going if you haven't done that yet.

After that, in this mod's config.php, set the force_voice_type setting to true. If you're installing this into an ongoing game, you'll need to save/reload and then zone transition to get all of the voice types updating.

[crazyduck]

1 hour ago, Min said:

 

The URL for that should be https://<your-pod>.proxy.runpod.net

 

Follow the mantella setup instructions for getting a mantella xtts runpod going if you haven't done that yet.

After that, in this mod's config.php, set the force_voice_type setting to true. If you're installing this into an ongoing game, you'll need to save/reload and then zone transition to get all of the voice types updating.

Ahh ok i have xtts run native in windows so your fix will not work right?

Edited September 15, 2024 by crazyduck

[Loopdeloop]

3 hours ago, Min said:

 

Also - FWIW, it may just be a matter of "This new release is more script intensive than the old one and it's pushing your papyrus setup over the limit" (Increased script load is mostly due to that script for npc context awareness I mentioned). I do see from the log that you're running a couple other really script heavy mods - Namely, "Wet and Cold" and XPMSE's styles script. That XPMSE script in particular is notoriously poorly performant. You don't need to run the XPMSE esp / scripts to use the XPMSE skeleton that mods require.

Thanks so much for the update and the tips. I removed those superfluous mods you mentioned (and a couple of others), then started a new game, to reduce script load but still had the same problem. Only removing MinAI_DISTR fixed it completely. 

 

Now I just need to test the mod to see if it works. Do the keywords work the same way as the Mantella version, in that I convince an NPC to say them?

 

Edit: Ah, I notice the prompts have changed a little? Eg 'sex' is now 'startsexscene'?

Edited September 15, 2024 by Loopdeloop

[Min]

3 hours ago, Min said:

 

I think the system to update context is putting too much load on your script engine in that version. I just pushed a small little release here: https://github.com/MinLL/MinAI/releases/tag/beta15 that improves performance a fair bit. See if that helps?

Beta15 has a serious bug - took it down for the moment. The rest of this still applies though.

 

You can also remove the MinAI_DISTR file that ships with the mod to disable the context effect, which is where most of the script load comes from. See if that helps, if beta15 doesn't.

 

Send me a copy of your papyrus.0.log on beta15 if that doesn't help. 

 

I spent a bit of the morning looking into this and found the bug that was causing performance and stability issues in Beta14+. It's fixed now in Beta17.

 

Beta17

 

• Fixed race condition that could cause mod failure. This impacted beta14 - beta16.
• Performance improvements
• Added ability to arrange for Carriages via natural dialogue

[Min]

36 minutes ago, Loopdeloop said:

Thanks so much for the update and the tips. I removed those superfluous mods you mentioned (and a couple of others), then started a new game, to reduce script load but still had the same problem. Only removing MinAI_DISTR fixed it completely. 

 

Now I just need to test the mod to see if it works. Do the keywords work the same way as the Mantella version, in that I convince an NPC to say them?

 

Edit: Ah, I notice the prompts have changed a little? Eg 'sex' is now 'startsexscene'?

 

Are you using AIFF?  The old Mantella version should work the same as before. The AIFF implementation is completely different. There are no more keywords, instead the LLM communicates actions outside of the main conversation text. Other than that, yeah, same general concept

[Min]

1 hour ago, crazyduck said:

Ahh ok i have xtts run native in windows so your fix will not work right?

This fix specifically lets you use Mantella's xtts configuration. The key difference is that AIFF sends a string like "whiterun_guard" to the xtts service, where-as Mantella sends "maleguard" (The base voice type). My fix forces AIFF to send "maleguard" so that the mantella xtts configuration can work with it.

Edited September 15, 2024 by Min

[billypnats]

Just switched to FF-AI. can you give some examples of how to start NSFW scenes? The narrator and AI follower seems to be aware of things If I manually start scenes with sexlab spells but can't seem to start it organically

[randomname002]

Sometimes you need to encourage them to "remember to use the correct actions" (if using AI-FF) or "keywords" if using Mantella

[wolfbread]

21 hours ago, Min said:

Sexlab AnimSpeed sounds like a cool optional dependency - Do you know if it supports VR? I'm likely to only integrate mods that can work for me.

https://www.patreon.com/posts/animspeedse-103787087

I don't know, is AE compatible with VR? 🤔

[billypnats]

26 minutes ago, randomname002 said:

Sometimes you need to encourage them to "remember to use the correct actions" (if using AI-FF) or "keywords" if using Mantella

I'm an idiot and put loose files into the ext folder instead of the folder. it's saying

• minai_plugin: Server/Papyrus plugin. Must also install the papyrus plugin

I'm not sure what plugin it's referring to, I installed all the requirements outlined on FF-AI website

[Min]

3 minutes ago, billypnats said:

I'm an idiot and put loose files into the ext folder instead of the folder. it's saying

• minai_plugin: Server/Papyrus plugin. Must also install the papyrus plugin

I'm not sure what plugin it's referring to, I installed all the requirements outlined on FF-AI website

That message indicates that the plugin is loaded. It's reminding you to also make sure to install the papyrus plugin.

[billypnats]

Working well now, but is there a way to disable the kiss and groping? THe animation isn't always appropriate depending on the setting