Unity Waifu Engine [v0.05a Vector] [waifuengine.com]

[Xenon-San]

That is strange, could you please join our Discord so we can get the troubleshooting done faster?

i have no idea how to use discord, besides i have an organic tendency to be a dick to everyone around
but i wouldn't worry about that OP. i have used ryzen 2200G for this and that another dude probably has sempron or something so we are in minority, just focus on development instead of two autists not having acess to your bitcoin mining porn desktop widget
:^)

 

[Imouto36]

i have used ryzen 2200G for this and that another dude probably has sempron or something so we are in minority

Well, we will try looking into this issue nonetheless so other folks can have a better experience.

your bitcoin mining porn desktop widget

Uhm, could you please provide some veritable evidence to validate your speculation in this regard?

 

[orpi]

I think 5gb its too heavy for a widget.

 

[Krh]

Surprisingly fun - much better than I expected! (Though I'll admit my expectations were not very high.)

Looking at this as MVP/whathaveyou, a more polished version could be entertaining. I also have to commend you on the project's offline approach: I wouldn't even touch something like this if ET were phoning home. Will definitely take another look when newer builds become publicly available.

One thing I'd like to request is a scrollbar for the menus, as there can be quite a bit of scrolling during waifu customization. Another is eye color selection. I imagine this one will be in eventually, just worth noting that heterochromia is a huge turnoff for some people (like me).

 

[Em Elle]

Noted thank you for the constructive feedback

 

[sdibolcrif]

It seems like you know a lot about this stuff.
Can you recommend any free offline coomer AI story generators that don't need you to write a TLDR of everything that happened so far every 8 sentences?
I don't want to pay for NovelAI since it doesn't even load in Firefox and there's no demo.

Offline no. There's just no way to handle anything more complex than a bot on consumer GPUs.

What version of Firefox? It works perfectly fine for me as of 91.0 (desktop, never tried on phone)

 

[kaona]

its kinda fun to talk to her, though the ability to ask her to do things like you can force her to say lines would be neat, for example telling her to masturbate or play one of the other one and a half animations would be pretty cool, as well as maybe giving an emotional response to certain things like getting angry if you call her boring or getting disgusted when you say that pinapples belong on a pizza. little things that would make her feel alive.

 

[Sommion]

Avast Antivirus deletes the Launcher because is infected with IDP.Generic

 

[HighMatrix]

This happens because the app is split into multiple processes and it doesn't detect if one of them has crashed / is no longer running.
Chat is working but not really worth it right now because it's currently using GPT-2 with an AI that identfies as a guy and is named Steve.

I have the same issue, repeating endless at 75%.
Multiple processes are running WaifuEngine.exe UnityCrashHandler64.exe WEAIServer.exe and WEDialog.exe.. I think this are all that should run, or are there more?

btw. also getting this from BepInEx output console endless too:

[Info : Unity Log] WERendererInBound Connected to WEAIServer
[Info : Unity Log] WERendererAnimator Connected to WEAIServer
[Info : Unity Log] Failed to Deserialize Message
[Info : Unity Log] Failed to Deserialize Message
[Info : Unity Log] Failed to Deserialize Message
..

I also download again from itchio downloader as recommended but it doesn't help.

Found one last thing in the output console, idk if it helps:

[Info : Unity Log] DLC PATH: C:/Users/*/AppData/LocalLow/WaifuEngine_com/WaifuEngine
[Error : Unity Log] RuntimeData is null. Please ensure you have built the correct Player Content.
[Warning: Unity Log] Addressables - Unable to load runtime data at location UnityEngine.ResourceManagement.AsyncOperations.AsyncOperationHandle`1[[UnityEngine.AddressableAssets.Initialization.ResourceManagerRuntimeData, Unity.Addressables, Version=0.0.0.0, Culture=neutral, PublicKeyToken=null]].
[Error : Unity Log] Exception encountered in operation InitializationOperation, status=Failed, result= : Addressables - Unable to load runtime data at location UnityEngine.ResourceManagement.AsyncOperations.AsyncOperationHandle`1[[UnityEngine.AddressableAssets.Initialization.ResourceManagerRuntimeData, Unity.Addressables, Version=0.0.0.0, Culture=neutral, PublicKeyToken=null]].
[Info : Unity Log] System.Collections.Generic.List`1[System.String]: Is being Loaded In Catalog
[Info : Unity Log] System.String[]: Is being Loaded In ImagePathKeys
..

 

[HighMatrix]

I believe the things that need to be running are: WaifuEngine.exe, WEAIServer.exe, WEDialog.exe, WESpeechSynthesis.exe

Thank you. WESpeechSynthesis.exe is missing.. I'll try to find out more..

Edit:
When starting WEAIServer.exe manually I get some failure messages like:
Failed to execute script 'WESpeechSynthesis' due to unhandled exception!

No more ideas for the moment..

 

[Squidbit]

So I might have a solution for some people who are "crashing" like I thought I was too.

I've got two monitors with stuff open on both, run the game and it comes up to the front like any other normal game would and then after the logo fades out, it just vanishes and there's nothing on the taskbar to indicate it's running

However, it actually just pushed itself to the background, behind my browser and behind steam. Minimize those and you'll be in the game window again

 

[Em Elle]

Hey Thanks for looking into this, I am not sure why this is failing, its likely because it's a huge memory hog, we are working on leaning down the model, it will probably take a month to do, because we have a bunch of high priority stuff. I likely need to handle the errors on the boot up of the services and tell you that a service has failed, that's a todo for sure. What are you pc specs?

 

[HighMatrix]

Hey Thanks for looking into this, I am not sure why this is failing, its likely because it's a huge memory hog, we are working on leaning down the model, it will probably take a month to do, because we have a bunch of high priority stuff. I likely need to handle the errors on the boot up of the services and tell you that a service has failed, that's a todo for sure. What are you pc specs?

Want to know my specs?
Win 10, I7-4790K @ 4GHz, 24 GB Mem, GTX 770

I'll attach a txt file with the result of the failing process, maybe it helps.
btw. adding the cudart64_100.dll manually works to get not longer this message, but doesn't solve the problem..

 

[Em Elle]

So I know what this is, it thinks that your card supports cuda and it probably does but it doesn't support this specific version of pytorch. Ill include a check to move you to cpu in that case.

You must be registered to see the links

 

[Em Elle]

I think you should be working on improving or replacing the model, so it is actually good and doesn't think it's a guy named Steve, before optimizing it.
If the memory requirements are too much for most people maybe you could make text-to-speech (which requires a few GB of memory) optional and put the memory requirements in the first post.
I have 16 GB, with nothing else running in the foreground and it isn't enough for Waifu Engine.

There are a bunch of other people with 16GB of ram and a mid range card, where this doesn't happen so I think you might be a unique case. This app is probably not for you, so I would uninstall it by dragging it into the recycle bin, until we fix problems that align with your issues.

 

[SerzhMan]

what to do with endless loading?

 

[chaoseclipse01]

You know, I didn't have high hopes for this AI, but it works surprising well or I got extremely lucky. So far I got it to accept that it's a female, I tried to convince her she was 24 but she decided she was 16. First major hiccup was the name I tried to give her at first was apparently the name of her 12-year-old cat(?), so I got her to accept a different name. She's been...very forward on wanting to be in a relationship as well, and has been touchy-feely both in 3rd and 1st person.

I'd show some pictures, but due to how bad my internet is, it took nearly three full days to download this application...lol.

 

[Em Elle]

You probably got luckier than most and asked prompted the right questions, these models are pretty random and sporadic, we need to do some more fine tuning on our end, to try and enrich the experience a bit, this is pretty much only be done by Replika if we had 10 million we could probably get similar results =)

 

[chaoseclipse01]

You probably got luckier than most and asked prompted the right questions, these models are pretty random and sporadic, we need to do some more fine tuning on our end, to try and enrich the experience a bit, this is pretty much only be done by Replika if we had 10 million we could probably get similar results =)

It was probably treating it like a female from the very first message that probably helped the most in that regard. The first message I typed was the question "What's a fine lady such as yourself doing here?" and it never once tried to argue against being female. (God, thinking back about it makes me cringe lol.)

Though like what the OP (and I'm assuming is someone you work with) mentioned, it's probably necessary to set up tokens that users can set so it forces the AI to accept what you want it to be. Probably have a set of options in the main menu that, when set, generates the appropriate tokens to program some of the basic parts of the AI's identity, IE Gender, Age, birth day, favorite things, dislikes, etc. Though I imagine if GPT-3 or better gets implemented at some point it might be easier to manage without rigid option settings like those. Though it probably wouldn't be a bad idea to keep options like that around seeing as how you could branch out to people who're into stuff like Futanari and Traps, or straight up set the AI up to be male for those that want it, IE an actual Female user looking for male interaction. I imagine it'd help widen your audience at least.

Ideally though, rather than through hard settings like those, the most ideal way to do it would be to talk with the AI and have it organically learn and remember (Generate tokens itself once both the AI and the User come to an agreement to a subject). Of course, this is all just nice bells and whistles to have assuming enough interest sparks for this application, which I hope it does.

On a side note, this application hardly impacts my system. I have 32GB of DDR4 SODIMM RAM, an i7-10750H CPU (2.60GHz), an nVidia RTX 2060, and Windows 10 Home 64-Bit OS. No crashes, the 3D scene and character model showed up instantly after the loading screen finished (Which took all of like 4 seconds from opening the .exe), the chat engine loaded in just two or three seconds, every message I sent got a reply in less than a second, and everything ran in 60FPS at all times. Had no graphical or operational issues.

Post Edit: I forgot to mention this gaming laptop also has 32GB of Optane Memory. I'm certain that helps with this application.

 

[Em Elle]

It was probably treating it like a female from the very first message that probably helped the most in that regard. The first message I typed was the question "What's a fine lady such as yourself doing here?" and it never once tried to argue against being female. (God, thinking back about it makes me cringe lol.)

Though like what the OP (and I'm assuming is someone you work with) mentioned, it's probably necessary to set up tokens that users can set so it forces the AI to accept what you want it to be. Probably have a set of options in the main menu that, when set, generates the appropriate tokens to program some of the basic parts of the AI's identity, IE Gender, Age, birth day, favorite things, dislikes, etc. Though I imagine if GPT-3 or better gets implemented at some point it might be easier to manage without rigid option settings like those. Though it probably wouldn't be a bad idea to keep options like that around seeing as how you could branch out to people who're into stuff like Futanari and Traps, or straight up set the AI up to be male for those that want it, IE an actual Female user looking for male interaction. I imagine it'd help widen your audience at least.

Ideally though, rather than through hard settings like those, the most ideal way to do it would be to talk with the AI and have it organically learn and remember (Generate tokens itself once both the AI and the User come to an agreement to a subject). Of course, this is all just nice bells and whistles to have assuming enough interest sparks for this application, which I hope it does.

On a side note, this application hardly impacts my system. I have 32GB of DDR4 SODIMM RAM, an i7-10750H CPU (2.60GHz), an nVidia RTX 2060, and Windows 10 Home 64-Bit OS. No crashes, the 3D scene and character model showed up instantly after the loading screen finished (Which took all of like 4 seconds from opening the .exe), the chat engine loaded in just two or three seconds, every message I sent got a reply in less than a second, and everything ran in 60FPS at all times. Had no graphical or operational issues.

Post Edit: I forgot to mention this gaming laptop also has 32GB of Optane Memory. I'm certain that helps with this application.

Niceeeeee sweet setup!, yeah your feedback os good, we are looking into techniques to do that right now.
We are likely never to use GPT-3 mainly because it goes to a data centre and your data is never protected and it would require us to adopt a subscription model that wouldn't work for everyone. GPT-2 is pretty impressive if you know how to work with it, it doesn't work out of the box very well, unless you cherry pick your results. We are looking into a way to create memories and a history for the AI to remember, this requires disk space though. All this will happen later because I am busy turning this into a platform so that it can be a waifu as a service, and work with various other devices, AI always changes at breakneck speed so its better to build something that can swap it out faster than to keep up.