AnotherMike

Member
Mar 26, 2020
161
132
208
Huge thanks:1.5, man, for all the hard work.;)
I'm a Asian lover, so I tweaked a bit in your prompt file so that the tan skin is now fair skin Asian, and the black skin is now tan skin Thai.
Also made some adjustments in full prompt format and "SD_get_position_prompt" area(ideas from
ninjaboi4's file), together with some minor adjustments. Now the guy will automatically appear in the image when sex happens, and I have added different view angles for different positions of sex.
I will attach the tweaked file here for any other Asian lovers.
(There are some problems with girls in the gaming cafe and Vendenberg Ltd. can not generate image)
That's a cool way to adapt it for personal preference!

For the bug with the gaming cafe and Vandenberg Ltd., I just updated my previous post with a fix. It's the SD_get_location_prompt function in the promptbuilder, in case you want to copy the fix to your version.

It seems that the location names caused issues, while I have no idea why the dot in "Ltd." caused issues when I've used dots in other parts of the prompts, I've just acknowledged that they do (or at least that seem to be the cause) and changed the prompts for those locations, I've also changed it for a bunch of other rooms with suspicious names just in case.
 
  • Crown
Reactions: Joshego

Joshego

Newbie
Donor
Aug 7, 2016
61
51
208
That's a cool way to adapt it for personal preference!

For the bug with the gaming cafe and Vandenberg Ltd., I just updated my previous post with a fix. It's the SD_get_location_prompt function in the promptbuilder, in case you want to copy the fix to your version.

It seems that the location names caused issues, while I have no idea why the dot in "Ltd." caused issues when I've used dots in other parts of the prompts, I've just acknowledged that they do (or at least that seem to be the cause) and changed the prompts for those locations, I've also changed it for a bunch of other rooms with suspicious names just in case.
The error is very much fixed, cheers!
Since you are using a RTX gpu, I don't know if you know the extension "TensorRT":
It will make image generation about 50%-80% times faster with RTX GPUs.
 

madeofcrust2

Newbie
Nov 18, 2018
19
4
141
So I missing Ophelia, Candace, Erica, Kara and Kara's mom. Do they come later in the game, since I'm on Nov. 11, in the game or are they suppose to show up earlier?
Their events trigger when you reach the Tier 1 of serum research (I think). If you use the Cheat tools to unlock all serum tiers (in the the same cheat menu where you can add to your company more funds of supply) then their events will not fire and they will not appear, it happened to me the other day.
 
  • Like
Reactions: lan1967

trkjti

New Member
Apr 6, 2017
12
3
22
I don't really know. As far as I know, the first version of the mod came from someone on Discord, then Ycbelane on these forums added img2img support and other things to that mod, then I used that version as a foundation for this one and at the same time, other people started creating their own versions, I don't know if anyone along the way created a git for it.

Personally I wouldn't mind if someone more dedicated to the project than me created a git based on this version but honestly, I think the original author of the mod from discord is the one who "owns" the code to the largest extent and should have some kind of final say about anything being created (if they haven't already created one?)
Thanks for the answer
I have asked on discord I'll see if I can any answer
 

Tad Cooper

Member
Feb 16, 2018
157
261
239
Appreciate the response. I figured there was a requirement but I tested the wrong variables. I know that some responses are based off of whether a trait is there or not. I would have checked the if then statements but I'm not sure how get into the code of the game yet.
Sure thing. It's easiest to pull from the . Don't be dumb like me and use Unren. This game heavily depends on default arguments and Unren can't supply them so you'll get crashes left and right complaining about missing variables. Big hassle.
 

Darkmind

New Member
Jan 27, 2019
12
9
30
Just wanted to follow up on my experiment of running a recent Windows build of Lab Rats under Crossover with the AI mod and StableDiffusion being run directly in Mac:

It all works, fairly straightforward. I needed a slight adjustment of the StableDiffusion command line from what the ReadMe suggested, but that was suggested by StableDiffusion itself so it wasn't a concern. (And may be specific to my hardware.) Feels a bit clunky to get running, but it all works once everything is spun up.
 
  • Like
Reactions: Bluecats

bustanv

New Member
Oct 10, 2020
2
1
14
The error is very much fixed, cheers!
Since you are using a RTX gpu, I don't know if you know the extension "TensorRT":
It will make image generation about 50%-80% times faster with RTX GPUs.
You saved me A LOT of time with this info. Thanks!
 
  • Like
Reactions: Joshego

lan1967

Member
Dec 15, 2017
251
68
213
It is the police chief, Christine. One you get her stats up high enough, the option will be available in the policies.
somehow I broke the game and the option does not appear. Also, I started to get constant error message on each time pass. Naomi Walters did not appear. Unisex event did not happen.
 
Last edited:
Apr 23, 2018
282
82
232
I'm not sure why but for whatever reason clarity costs are through the roof. For a first time using strengthen opinion game wants to charge me 1600 clarity. It's supposed to start and 200 and go up from there.
 

trkjti

New Member
Apr 6, 2017
12
3
22
I don't really know. As far as I know, the first version of the mod came from someone on Discord, then Ycbelane on these forums added img2img support and other things to that mod, then I used that version as a foundation for this one and at the same time, other people started creating their own versions, I don't know if anyone along the way created a git for it.

Personally I wouldn't mind if someone more dedicated to the project than me created a git based on this version but honestly, I think the original author of the mod from discord is the one who "owns" the code to the largest extent and should have some kind of final say about anything being created (if they haven't already created one?)
Original author replied on Discord. Je did setup a git here :
I didn't check if some of your modifications were included.
But it looks like he is willing to accept contributions
 

TheItch

Newbie
Jul 24, 2021
44
162
124
I'm not entirely sure what you mean with pre-prompt capabilities? are you talking about stuff like the "score_9, score_8_up, masterpiece, high detail" and so on you often see at the beginning of prompts? There's a setting called "Prompt style" that is attached to the end of the prompt where you put things like that. In the version of the mod attached to this, I've moved the style prompt to the beginning of the prompt since most people seem to want to put that kind of parameters in the beginning of the prompt.

Other than that, I agree that it doesn't follow the actions perfectly, and the truth is, I doubt it ever will. With that said, I've included an updated version where I've made a bunch of changes, which at least improves the possibility for the user to pick and chose which parts they want to compromise with.

Other than what I've added in previous versions, this new one includes the following:


1.
better prompting for the sex positions, I'm mostly using SD models based on the pony model in my testing so these prompts are focused on working with such models, though even if the SD models understand the prompt, there's still no guarantee that they will always honor the positions requested in the prompts (and there are still some positions where I have absolutely no Idea how to write them for a good result).

2.
I've also added a setting called "use basic prompts for positions" that completely disable my prompts for sex positions and instead go back to the old positions that are a simple descriptions of the images used, you'll see why that might be useful in the examples a bit later in this post.

3.
I've included cum in the prompt, I think I hadn't added this in the previous version, either way, if the girl has cum anywhere on her, it is now added to the prompt.

4.
A setting for each girl called "change name to:", which changes the name of the girl (only in the prompt) to whatever you enter there. You could use this to change the girls appearance accordingly which helps a lot with consistency in their appearance. Just make sure you only enter the name of fictional characters there and don't put in the name of any celebrities or other real people since that'd be highly immoral :p Strictly speaking this setting wasn't needed since you could already do this with the "append:" setting but I preferred to have it cleanly as it's own field (and not have multiple names entered in the prompt since some of the "original" names might get recognized by the AI).

5.
I've moved the "Style" prompt to the beginning of the prompt, I'm not sure how much of a difference it makes, but most example I've seen whether it's actual styles or the "score_9, score_8_up..." often used for pony models, people always seem to put it at the beginning and at least chatGPT claims that the early parts of the prompt might more strongly influence the result :) I've also made other changes to the order things are placed in the prompt based on what I think seems best.

6,
I've also added a negative style prompt, you could use it for the typical "score_6, score_5, score_4" found on pony models as well as reinforce something like "realism" in the positive prompt by adding "cartoon" in the negative prompt as well as other things. There's already a bunch of hardcoded things put into the negative prompt, I've left them there for now but maybe they should've all been removed to allow the user to pick their negative prompts more exactly?

7.
I've added a "male description" prompt, if you do use my more explicit sexual position prompts, this will be added at the end of the prompt. Here you could put a description of the male character that appears in the sex scene.

Though it's important to notice that stable diffusion really suck at separating descriptions between characters, so if you put something like "muscular white man", it's quite possible that the girl in the scene will become muscular and white as well. I've seen some people use "BREAK" between the male and female descriptions to separate them but I have no idea how useful that actually is, but sure, if you want, you could put something like "BREAK, BREAK, BREAK, BREAK, white man, large penis" in the male description if you like :)

honestly, I've had very varied results with this one so don't get your hopes to high for it.


8. I've fixed the "watcher" situation. When someone was watching you having sex, the watcher would be shown in the same sexual position as the girl you were having sex with, now she should behave more in line with the expectations.

There's probably more that I've forgotten and some of these changes are fairly new additions so I haven't exactly tested them extensively, use at your own risk.

For some reason the style, negative style and male description prompts sometimes seem to "vanish" and I haven't been able to figure out why those settings are sometimes lost. If you write elaborate long texts in either of those prompts, you might want to save them separately in a text-document so you can easily copy & paste them back into the game just in case they vanish between gaming sessions.

As always, if you want to use the background remover function, remember to install the ABG Remover extension:

View attachment 4951837


Now then, some examples and considerations when it comes to the different settings:

If you use img2img, I highly recommend setting a low denoising strength (somewhere around 0.3-0.5 seems good) while also using the "Use basic prompts for positions" settings. If you do that, you'll basically have the AI improve the originally images but keeping all the poses and so on exactly the same, so it'll be as good (or as bad, depending on how you see it) as the original game when it comes to showing what is going on.

These are examples of blowjob, missionary, doggy style and anal pile driver with such settings:
View attachment 4951841

-----------------------------------------------------------------------------------------------------

If you want it a bit more advanced, but still keep things reasonably easy for Stable Diffusion to follow, you could try leaving the "basic prompt" setting on but disabling the img2img setting, then you'd still have just basic positions (lying on back, on all fours) and so on for the different sexual positions, but Stable Diffusion would be more free to improvise with how it's presented.

Once again, here are examples of blowjob, missionary, doggy style and anal pile driver with such settings:
View attachment 4951842

-------------------------------------------------------------------------------------------------------------------------

If you want to go completely wild,you can disable both img2img and the basic prompt settings, then you'll get more detailed sex positions and as mentioned above, at least most pony models will, to some extent, recognize the prompts but how well they respond to them will differ greatly. In my (limited) experience, slower SD models will work better here, but will require more patience.

Using something like Cyberrealistic Pony or Pony Realism with 5-7 CFG Scale and 25-35 steps has produced "somewhat" consistent results for me, but certainly still far from perfect and with such settings, it takes 20-30 seconds for each image to get generated on my 5070Ti (with ADetailer enabled).

Once again, here are examples of blowjob, missionary, doggy style and anal pile driver with such settings:
View attachment 4951840



As mentioned it still gives weird results now and then but it does a surprisingly good job following the prompts. Though the penis is a bit all over the place when it comes to shape and color and whether it's even connected to anyone or not :)

To fix that, there's the male description option mentioned above, putting "white man, giant penis" in that field, I got the following result:
View attachment 4951839

But I should point out I was probably lucky at this time since everything was responding almost perfectly to my prompts when gathering images for these screenshots. As mentioned previously, even with a "slower" SD model, it still gives the random weird results now and then:
View attachment 4951850

Especially piledriver position seems to be something all SD models struggle with, to the extend I've almost considered replacing the prompt for it with something easier for Stable Diffusion to generate.

Sometimes it seems to really get "stuck" on bad/weird positions, like in one example I had the doggy style sex look completely messed up and even tweaks to the settings, switching between anal/vaginal or removing clothes did not help, the girl would always be twisted in a weird angle, it almost seemed like the "seed" was just bad for that position, in those cases, just open the settings, pick "generate new seed" and try again, in my case that fixed it.

Sadly, for the people reporting getting all white backgrounds with the background remover, I haven't been able to reproduce that bug and I'm kind of out of ideas as to what could be the source of it so most likely that bug is still there (unless one of my other changes just randomly fixed it but that seems highly unlikely).

Most likely this will be my last release of the mod. There seems to be so many different people working on different version that I'd rather see them get joined at some point than have the different versions branching out in different directions and since I'm incredibly lazy, I'm hoping someone else will take whatever they find useful from my version of the mod and incorporate into their own :)

Edit: Hasn't been long since I said it was the final version, but found a bug where it would break if you had backgrounds visible and entered the Gaming Café, seems stable diffusion (or at least the Automatic 1111 implementation of it) doesn't like the é character. I've fixed this specific instance but a more proper search and replace for "bad" characters should probably be created.

Edit2:
Ok, just ONE more bugfix, noticed another room causing issues when background remover was turned off, so I've now gone through all the rooms I could find with "suspicious" characters in their name and changed their description in the prompts. Proper "prompt cleaning" could probably still be needed in case strange characters might appear in other parts (character names?)
Maswoo set up a git repository today
I hope you don't have any commitment issues :D

EDIT: nvm, trkjti beat me to it XD
 
  • Like
Reactions: trkjti

AnotherMike

Member
Mar 26, 2020
161
132
208
Original author replied on Discord. Je did setup a git here :
I didn't check if some of your modifications were included.
But it looks like he is willing to accept contributions
Maswoo set up a git repository today
I hope you don't have any commitment issues :D

EDIT: nvm, trkjti beat me to it XD
But I do, I DO have commitment issues :p

Honestly, I don't know the first thing about python or ren'py, all my changes are a mixture of my knowledge from other programming languages and suggestions from chatGPT, I'm pretty sure it violates every coding-standard out there.

That combined with me not having ever used GIT (other than for downloading), plus, most importantly, my extreme laziness, means it'd be very much preferable to me if some helpful person would pick through my changes and only add the parts that might be considered useful, in a controlled and possibly better aligned with coding-standards way, to the GIT :)

Though, if that's not an option then I guess I'll have to look up how GIT commit works later on.

On another note, even though I said I wouldn't do any more changes, I just had an idea I had to try and after hours of fighting with ChatGPT that only suggested one thing after another that kept breaking the code, I finally manged to get it working, and it was almost COMPLETELY USELESS, at least in it's current form.

What I did was update the mod to interpret EVERY SINGLE LINE OF NARRATION (that is, the text in the game describing what happens, when it isn't someone talking) and pass it on to Stable Diffusion. If the narration says "She bends forward and jiggles her ass" then I send "She bends forward an jiggles her ass" in the Stable Diffusion prompt, if the narration says "What will you do know?" I send "What will you do know?" in the Stable diffusion prompt. And EVERY time, a new image is requested.

99.9% of the time this just results in Stable diffusion getting extremely confused since some times the narration isn't really describing what is happening (like variations "what will you do next?") or, it's part of a series of text where the context between each sentence is important and the meaning will be completely lost when just one, or a few, sentences get's sent to Stable Diffusion. Not to mention, most of the text uses various phrasing that makes it even more confusing (your meaty rod, etc.) and of course, stable diffusion doesn't have the slightest idea who this mysterious "you" is, maybe I should've replaced "you" with "man"?

Once in a blue moon it managed to pick up something useful from the narration, but most of the time, as I said, it just makes things worse while constantly hammering Stable diffusion for new images, since now, it tried to redraw things all the time, almost every single time the text changes (it only processes narration though, not dialog, so whatever anyone says won't trigger a new image).

This is controlled by a setting called "Process narration" and no one should ever turn it on :)

Here's a few times where it actually managed to get something useful from the narration, but as mentioned, most of the time it just makes things worse, but maybe someone willing to put more work into it might be able to build upon it and actually sort through those strings and look for useful pieces while discarding the less useful parts? :)


screenshot0178.png screenshot0176.png screenshot0173.png

I also made a less weird addition, there's something called "the_object" in the code which is the object you might be interacting with during sex (a couch, a bed, a desk, a windows, the floor and so on), I now pass that object along in the request. Though since I don't know if you're leaning against it (a wall, a window, a desk, a couch a table) or being on top of it (a bed, a desk, a couch, a table), I simply pass on the name of the object and leave if up to Stable diffusion to figure out what it wants to do with it.

Though at times I worry that adding more and more features like this is only making things worse. Even ignoring the silly narration part, the prompt is getting rather long and the more things you send in the prompt to stable diffusion, the more things it has to consider and the more difficult, and thus unlikely, it becomes for it to correctly fulfill the request (at least, that's how I understand it?)

And THIS time, it will almost certainly, maybe, be the last time I make any changes :)
 

TheItch

Newbie
Jul 24, 2021
44
162
124
But I do, I DO have commitment issues :p

Honestly, I don't know the first thing about python or ren'py, all my changes are a mixture of my knowledge from other programming languages and suggestions from chatGPT, I'm pretty sure it violates every coding-standard out there.

That combined with me not having ever used GIT (other than for downloading), plus, most importantly, my extreme laziness, means it'd be very much preferable to me if some helpful person would pick through my changes and only add the parts that might be considered useful, in a controlled and possibly better aligned with coding-standards way, to the GIT :)

Though, if that's not an option then I guess I'll have to look up how GIT commit works later on.

On another note, even though I said I wouldn't do any more changes, I just had an idea I had to try and after hours of fighting with ChatGPT that only suggested one thing after another that kept breaking the code, I finally manged to get it working, and it was almost COMPLETELY USELESS, at least in it's current form.

What I did was update the mod to interpret EVERY SINGLE LINE OF NARRATION (that is, the text in the game describing what happens, when it isn't someone talking) and pass it on to Stable Diffusion. If the narration says "She bends forward and jiggles her ass" then I send "She bends forward an jiggles her ass" in the Stable Diffusion prompt, if the narration says "What will you do know?" I send "What will you do know?" in the Stable diffusion prompt. And EVERY time, a new image is requested.

99.9% of the time this just results in Stable diffusion getting extremely confused since some times the narration isn't really describing what is happening (like variations "what will you do next?") or, it's part of a series of text where the context between each sentence is important and the meaning will be completely lost when just one, or a few, sentences get's sent to Stable Diffusion. Not to mention, most of the text uses various phrasing that makes it even more confusing (your meaty rod, etc.) and of course, stable diffusion doesn't have the slightest idea who this mysterious "you" is, maybe I should've replaced "you" with "man"?

Once in a blue moon it managed to pick up something useful from the narration, but most of the time, as I said, it just makes things worse while constantly hammering Stable diffusion for new images, since now, it tried to redraw things all the time, almost every single time the text changes (it only processes narration though, not dialog, so whatever anyone says won't trigger a new image).

This is controlled by a setting called "Process narration" and no one should ever turn it on :)

Here's a few times where it actually managed to get something useful from the narration, but as mentioned, most of the time it just makes things worse, but maybe someone willing to put more work into it might be able to build upon it and actually sort through those strings and look for useful pieces while discarding the less useful parts? :)


View attachment 4958519 View attachment 4958520 View attachment 4958523

I also made a less weird addition, there's something called "the_object" in the code which is the object you might be interacting with during sex (a couch, a bed, a desk, a windows, the floor and so on), I now pass that object along in the request. Though since I don't know if you're leaning against it (a wall, a window, a desk, a couch a table) or being on top of it (a bed, a desk, a couch, a table), I simply pass on the name of the object and leave if up to Stable diffusion to figure out what it wants to do with it.

Though at times I worry that adding more and more features like this is only making things worse. Even ignoring the silly narration part, the prompt is getting rather long and the more things you send in the prompt to stable diffusion, the more things it has to consider and the more difficult, and thus unlikely, it becomes for it to correctly fulfill the request (at least, that's how I understand it?)

And THIS time, it will almost certainly, maybe, be the last time I make any changes :)
Dude, that's some crazy scientist shit right there! :geek:
My $0.02 - You could try dividing the prompt builds into 2 categories - full scenes (so anything that is not narration) and "updates" (narration).
Keep the full scene until the next full scene shows up (refresh), but add all the the update prompts to the previous scene prompts with increased weight (i.e. wrapping calls in (()) brackets), raising the weight of each subsequent update and see what happens. Or, alternatively, only keep the most recent update with elevated weight.

But yeah, pronouns will have to be parsed, along with a bunch of other shit (i.e. dropping all updates that end in a question mark). But this way SD will have at least something to work with (and fuck up in a myriad of funny/grotesque ways)! :ROFLMAO:
 

AnotherMike

Member
Mar 26, 2020
161
132
208
Dude, that's some crazy scientist shit right there! :geek:
My $0.02 - You could try dividing the prompt builds into 2 categories - full scenes (so anything that is not narration) and "updates" (narration).
Keep the full scene until the next full scene shows up (refresh), but add all the the update prompts to the previous scene prompts with increased weight (i.e. wrapping calls in (()) brackets), raising the weight of each subsequent update and see what happens. Or, alternatively, only keep the most recent update with elevated weight.

But yeah, pronouns will have to be parsed, along with a bunch of other shit (i.e. dropping all updates that end in a question mark). But this way SD will have at least something to work with (and fuck up in a myriad of funny/grotesque ways)! :ROFLMAO:
Yeah, I should probably clarify that I do still keep the "main" prompt (or the full scene) as is through out everything, the narration part is just added on top of that for each call, otherwise it'd be even worse, but this is still enough to make Stable Diffusion really confused :)

The full prompt might be something like this (and I'd almost consider this sample a best case scenario) :
solo,grey eyes, happy, (tanned:1.5), dark skin, 20 years old teen named Emily Vandenberg in Coffee Shop, attractive, golden blonde twintails, white Bralette, dark teal Long Skirt, SFW, white Sandal Heels, large breasts, , (short body), (low stature), She rubs her cheek against your wet shaft., kneeling, (blowjob:1.5) , , floor

Most of the time that added sentence in the middle of everything will just be something confusing that seems to, at best, get ignored but when it's not ignored it's far more likely to make stable diffusion confused than help it get closer to what is actually taking place.

Adding up information from all the narration steps until the scene changes (or until someone says something, since I guess that'd be an interruption of the current narration) is definitely an interesting thought, though I'll leave that idea for someone else since the end result would still need so much processing before it turns into something useful :)

I think the best bet (other than hard code special cases for everything) would be to try and look for some common pattern in how things are described and only add things to the prompt when there's recognizable patterns (like if the words "rub", "cheek" and any euphemism for penis exists in the text, add "rubbing penis on cheek" to the prompt). but that would still be a huge undertaking to go through all of it. And to be honest, while it'd be the more realistic solution, it'd also lose some of it's charm, I like the idea of just sending everything to Stable Diffusion and say "here, work something out" :)
 

TheItch

Newbie
Jul 24, 2021
44
162
124
Yeah, I should probably clarify that I do still keep the "main" prompt (or the full scene) as is through out everything, the narration part is just added on top of that for each call, otherwise it'd be even worse, but this is still enough to make Stable Diffusion really confused :)

The full prompt might be something like this (and I'd almost consider this sample a best case scenario) :
solo,grey eyes, happy, (tanned:1.5), dark skin, 20 years old teen named Emily Vandenberg in Coffee Shop, attractive, golden blonde twintails, white Bralette, dark teal Long Skirt, SFW, white Sandal Heels, large breasts, , (short body), (low stature), She rubs her cheek against your wet shaft., kneeling, (blowjob:1.5) , , floor

Most of the time that added sentence in the middle of everything will just be something confusing that seems to, at best, get ignored but when it's not ignored it's far more likely to make stable diffusion confused than help it get closer to what is actually taking place.

Adding up information from all the narration steps until the scene changes (or until someone says something, since I guess that'd be an interruption of the current narration) is definitely an interesting thought, though I'll leave that idea for someone else since the end result would still need so much processing before it turns into something useful :)

I think the best bet (other than hard code special cases for everything) would be to try and look for some common pattern in how things are described and only add things to the prompt when there's recognizable patterns (like if the words "rub", "cheek" and any euphemism for penis exists in the text, add "rubbing penis on cheek" to the prompt). but that would still be a huge undertaking to go through all of it. And to be honest, while it'd be the more realistic solution, it'd also lose some of it's charm, I like the idea of just sending everything to Stable Diffusion and say "here, work something out" :)
Yeah, ironically that could be a decent prompt for Flux. That model loves long sentences. You'd have to be ok with waiting for 10-20 seconds on each image though XD (besides the other technical and otherwise finnicky stuff that comes with flux).
But you are right, making this work properly would require a stupid amount of work (realistically - baked-in back-end prompts for each text panel/sentence. And gooood luck with that! :D).

What i want to see (and have been trying to implement on my own) is forwarding sex pose information, so that i can properly prompt regular, static poses and separately - sex poses. This in and on itself would be an absolute MASSIVE change, as you would not be limited by the baked paper-doll poses any more and could unleash a torrent of appropriate prompts.
Then expand it to sex-scene updates (orgasming, slapping ass, creampie, w/e).

And also expose a lot more information about the state of the game and characters to the mod back-end to construct conditional prompts from. A lot more detail about clothes, transparency, underwear, visible cum, makeup, arousal... etc, etc.

There are so many great LoRAs that support sliders and could make each character and scene more unique (breast size slider, for example, allowing you to set precise breast size relative to checkpoint baseline, instead of relying solely on janky prompts), or LoRAs that allow you to modify clothing transparency, or quite literally - how deep the MC is penetrating a certain orifice (could be based on oral/anal experience, for example). The potential for granularity, precision and variety here is massive.

Aaand a prompt aggregator (aggregating and weighing prompts, based on how many conditions call each specific prompt), with an optional debug output, so that i can see how the prompts are composed and use it for debugging and tuning.

I have all of these ideas, but my coding illiteracy makes me feel like my hands are tied behind my back lol.

And the shit that maswoo and others want to implement goes waaay beyond all that XD

Anyway, you should join the LR2R discord. This is not the place for such chats. :)
 
4.60 star(s) 79 Votes