NaughtyAnon
Newbie
- Apr 2, 2020
- 44
- 36
- 152
Honestly the AI mod can do that if you really wanted. Just change the checkpoint you use. These are using milkyDreams.
NaughtyAnon
Newbie
- Apr 2, 2020
- 44
- 36
- 152
oh ok sorry i misunderstood, i thought the auto gen thing was another mod, ty the same
I'm using the same checkpoint and lora mentioned in this comment:
https://f95zone.to/threads/lab-rats-2-reformulate-v2025-04.32881/post-17374451
(to save a few clicks-- checkpoint:
You must be registered to see the links
- lora:
You must be registered to see the links
)Just for added clarification, I have an 8gb 3070 and it takes me around 8-10 seconds to generate each image (with 512x768 dimensions.. I haven't tried smaller yet). I noticed as I progressed through the game a bit that the genAI mod will often create images incorrectly and require a ton of additional prompt editing to dial back their inconsistencies. Two main examples:
1. - Every single image generated of Stephanie for me would make her topless and most seeds were giving her a fat belly. I suspect this is because the auto-generated prompt is adding (Overweight:0.41), but I never did figure out why she's always generated half-naked. Every seed I tried was the same result, and often she would also be missing pants. This is day one, the first time I'm meeting Stephanie, with no custom prompt edits and Override set to Off (this seed isn't making her fat, thankfully..lol):
2. - Every single image of Jennifer (MC's mom) would generate with two of her side by side, no matter which position I tried. I'm fairly confident this is because the auto-generated prompt states "1girl" then has a line break ("BREAK"), then "1 white milf" and several empty tags at the end ", , , , ". Based on my testing/editing it seems like the addition of these "BREAK" and ", , , , " lines is causing the checkpoint to think these are separate conditions or something, I'm not entirely sure. But I can say with confidence that manually overriding the prompt and removing those extra "BREAK" and ", , , , " consistently resolved my issues with this character. But again, that's not ideal... because that means I'll need to manually tweak the prompt every...single...time... that her character appears on screen, for every single unique interaction, in every single position... etc.
Personally I think this is a really cool mod, but it slows down game progression so dramatically for me since I end up disliking a lot of the images that are generated and "reroll" them by manually editing prompts so frequently that it's just not an ideal solution. I think this is more of a novelty of clever implementation than a real replacement for the original game images, at least in its current form.
It would be great if we could generate image sets for each character in advance, but since these images are all generated based on contextual information during each scene it's not something that can easily be done. I'm not really sure there's even a possible solution for this except to play through the game and manually spend 5-10 minutes during each scene tweaking prompts and regenerating images until you've got everything set up how you want it.
I do notice that images generate very reliably if you just use the img2img plug-in, since the mod will simply take the original game image of a character (including their clothing) and create a 1:1: replacement in the same style as the original. This may be the way to go if you want consistency and you don't want to have to spend 5-10 minutes during every scene manually tweaking prompts and generating new test images repeatedly until you get it right. But personally, I think going that route kinda lessens the genAI mod a lot and removes most of the fun/exciting stuff it can do. Here's an example of Stephanie with my checkpoint/lora though, using img2img (with background removed) for those curious:
Ultimately, I think everyone's experience with the genAI mod is largely going to be determined by how much effort they, the players, put into it. If your desire is to simply drop the mod into the game and never change a single setting, then I think you're going to be pretty disappointed. If you are fine with spending a lot of extra time tweaking each character and scene manually, then you're probably going to really love the capabilities here. And I've only tried two separate checkpoints (both listed in this thread) so maybe a different checkpoint would alleviate most of my complaints, not sure.
It's a bit too much work for me, sadly. Takes me out of the gameplay experience when I feel like I'm developing and troubleshooting the game rather than playing it, like I'm in dev-mode rather than gamer/gooner-mode, if that makes sense. Lol
----------
tl;dr: My review of the genAI mod is that YMMV. It can be amazing if you put in the work! But if you want this to be a fast one-click solution, you're gonna have a bad time.
EDIT: Fixed my post link at the beginning of this comment. I linked to the wrong post originally
Last edited:
Furthermore, using stable diffusion is like using a nuke to catch goldfish--bread crumbs are cheaper and easier
And by that, I mean the modders simply choosing a reasonable range of outfits, and then pre-generating them with AI for each named character and each situation
20ish named characters
10ish outifts
20ish poses
a few thousand images, and once it was successfully completed, you could choose a art new style and make a new image pack as an alternate download
or do only the sexual images, leaving the default for everything else, that would be half as much work
You'd have to throw away the generic characters, but who cares?
And by that, I mean the modders simply choosing a reasonable range of outfits, and then pre-generating them with AI for each named character and each situation
20ish named characters
10ish outifts
20ish poses
a few thousand images, and once it was successfully completed, you could choose a art new style and make a new image pack as an alternate download
or do only the sexual images, leaving the default for everything else, that would be half as much work
You'd have to throw away the generic characters, but who cares?
That'd probably be in a spin-off mod or something. The current storyline doesn't really support it since the MC's character shown in his expositions and interactions are generally consensual/perverted/gray. Adding the rape/serum thing would make the character psychotic (flip flopping between feeling bad about pushing incest / corrupting his family etc and outright raping other women while maintaining a legal pharmaceutical empire)
This is partly what killed Haramase Simulator since it got to the point people were too confused to fap
The only compromise I can think of is the slave play / dungeon / BDSM club / adding of a new Female NPC that likes rape roleplay
But say, making MC suddenly decide to rape Stephanie or Alexia out of nowhere would literally break the game both coding wise and story wise.
Try using a denoising strength of 1.0 with img2img. That will give the most freedom of interpretation to the AI from that starting image.
About the (overweight:0.41)...
Prior note: I'm new to AI image generation, so I'm not the most reliable knowledgeable person to give tips, but if I understood correctly what I've been reading elsewhere (and is consistent with the results I'm getting) i think there's a big misunderstanding to what the GenAI modders/creators intended.
Let's analyse these 4 options of overweight and what they mean: (overweight:0.25) / (overweight:0.50) / (overweight:0.75) / (overweight:1.00)
The modders idea was to create a linear progression in someone's level of overweight, from slightly overweight to overweight - clever and simple. But if my readings are correct (and my own testings seem in line with those) what you are saying in the prompt is that the word overweight should be more or less considered in the ideas contained in the prompt. It's a level of importance of the word within the prompt not a level of overweight in the generated image. If taken into consideration in the prompt evaluation by the AI, the word overweight will likely have the same meaning regardless of the number that follows it, what changes is the amount of times the AI will use it as description for the image being created.
This is not easy to explain, I hope the message got through and it helped.
Yeah, makes perfect sense really, and it largely lines up with my (albeit rather limited) experience when working with LLMs and editing some of my own simple chatbots. Based on my username you can probably infer this, but I played a lot of Adult Interactive Fiction games (AIF, or text adventure games) back in the day when they were more popular, and someone clued me into the fact that you can use modern LLM chatbots to largely recreate those types of gameplay experiences. One of the things I've noticed while editing LLM chatbot prompts is that phrasing and terminology is extremely important, and usually what you leave out is more important than what you include.
---
Negative prompting is a great example of this. Let's say you want to make a chatbot that discusses space travel but you want to ensure they never bring up the topic of politics. How do you do this? The first thought here is usually to include a qualifying statement somewhere in your prompt that says something to that effect: "politics should never be discussed" or "under no circumstances should politics be mentioned during this chat." Sounds fine, and in most cases the LLM will recognize the intent here and steer clear of that topic, but often the model will still bring it up simply based on the fact that 'politics' is in the prompt at all.
As I understand it, in a way it's kind of like the Streisand effect. That we've included it now means that the LLM must 'consider' that term in its responses and weigh whether or not it should be mentioned. The qualifiers of "never" and "under no circumstances" then become almost secondary to the term "politics", and we'll see way more responses including it than if we'd have simply left those statements out entirely.
Of course this isn't always the case, but it happens often enough that I'm confident it's a thing. Regarding the overweight issue above, I believe one other reasons it's focusing on that trait is that Stephanie's auto-generated prompt is actually quite short and I think that's leading the LLM to only consider those few traits (EDIT: In hindsight, I'm now beginning to think this is why she usually shows up half-naked too). To take that a step further, if I were to request an image of "woman with brown hair, (overweight:0.1)" and absolutely no additional details, I really shouldn't be surprised to find that most of the outputs will be of an overweight woman with brown hair. Weighting (the ":0.1" part) can be extremely useful, but a lot of times if there isn't enough information or other details the model might still boost its importance in the prompt.
---
Unfortunately, based on the way Lab Rats 2's character traits work, I'm not really sure there's an easy solution here, or even any real solution at all. We're fighting the checkpoints to try and extract a reliably precise image from a small amount of information, and to be honest the mod already does a phenomenal job of this (hats off to the folks who built it! it really is awesome) but with the way LLMs work this type of thing is always going to be an issue.
Still, this is all pretty new and I'm hopeful some clever folks will come up with a more reliable solution as the genAI mod is iterated. And there's still a few people working on the DAZ rework, at least from what I've seen on the Discord. Lots of stuff to look forward to yet!
I'll try increasing the denoising strength to 1.0. I set it to 0.5 previously and noticed I was getting somewhat better images but clothing that didn't always match what the character was actually wearing. Again, one of those trade-offs to be made... accuracy, or freedom to let the LLM be creative?
Last edited:
Silverdragon
Newbie
- Mar 22, 2018
- 67
- 55
- 100
I know I probably could find the answer through search, but I will ask anyway.
Does this version have more content than the original, or is it still worth playing the abandonned version?
Does this version have more content than the original, or is it still worth playing the abandonned version?
IMO, not worth playing the abandoned version. Modded has a lot more content and is a strict upgrade to my knowledge
I can't get this to work at all sadly, it's so insanely complicated and as soon as one thing breaks it's like ... what to do?
When I'm in LR2 the connection to SD just doesn't work, and the only thing I'm not doing is running the Stable Diffusion webui with the arguments included in the readme (set COMMANDLINE_ARGS=--medvram --no-half-vae --listen --port=7860 --api --cors-allow-origins *) because if I do that the webui doesn't launch at all.
When I'm in LR2 the connection to SD just doesn't work, and the only thing I'm not doing is running the Stable Diffusion webui with the arguments included in the readme (set COMMANDLINE_ARGS=--medvram --no-half-vae --listen --port=7860 --api --cors-allow-origins *) because if I do that the webui doesn't launch at all.
User_68P
Member
- Jul 17, 2023
- 194
- 2,624
- 376
The commandline stuff is what allows the LR2 client to connect to the Stable Diffusion server. Without them it will not work.(Medvram and no half are actually not a must, but they are normally good options to use).
But there are so many things that could be wrong (as you say) depending on your system, both hardware and software. If you are a Discord user I would recommend to join the LR2 Discord. People are asking daily on these topics there, so you might find solutions and help, and there is a channel dedicated to the genAI mod.
But there are so many things that could be wrong (as you say) depending on your system, both hardware and software. If you are a Discord user I would recommend to join the LR2 Discord. People are asking daily on these topics there, so you might find solutions and help, and there is a channel dedicated to the genAI mod.
Is there a list of commands for consola somewhere?
How do I edit if a particular girl is pregnant?
It's not in cheat mod, it's not in URM, it's not even in saveeditiononline.
Besides, when I run the realistic pregnancy system most girls immediately change the fertility setting to IUD.
I am using the latest release as of July 20, 2025.
How do I edit if a particular girl is pregnant?
It's not in cheat mod, it's not in URM, it's not even in saveeditiononline.
Besides, when I run the realistic pregnancy system most girls immediately change the fertility setting to IUD.
I am using the latest release as of July 20, 2025.
Baudler
Member
- Nov 22, 2021
- 126
- 97
- 128
It really should use a lora with variable strengths. I used to mess around with PonyXL but if you wanted to
You must be registered to see the links
you have to use a lora. And even then it's tricky, at least with stable diffusion type models because image aspect ratio / camera angle interact with it.For example, this is a good lora for varying weight:
You must be registered to see the links
It only breaks down at the extremes (beyond just fat).
Burt
Ignoring Social Media
- Dec 14, 2016
- 1,315
- 1,023
- 442
Playing the original would help you understand how good this one is though but to be fair to the Original Dev good idea's just not the skill to carry them out
This version has more content, but also changes to its mechanics and inconsistencies in tone and how characters are written. Whether you prefer this or the original more is really down to personal preferences. Try both and see for yourself
Garybaldi_the_Vile
Member
- Apr 28, 2018
- 185
- 156
- 258