Mod Lab Rats 2 Reformulate GenAI Mod Thread

[World of Sins]

Have you ever tried Pony v14.1 with DDIM CFG++? It works pretty well.

 

[The Huntsman]

*facepalm* You are correct. Those are all in a separate subfolder I completely skipped. I'll work them in over the next few days to test.

Hey, you seem to be the best expert I can find on here for creating your own prompts... Do you have any advice on how to get started with actually editing the prompt builder? I don't understand a lot of the language in it and so it's hard for me to tell what to change and what to leave alone, and what doesn't even matter...

Any advice would be incredibly helpful!

 

[NaughtyAnon]

Hey, you seem to be the best expert I can find on here for creating your own prompts... Do you have any advice on how to get started with actually editing the prompt builder? I don't understand a lot of the language in it and so it's hard for me to tell what to change and what to leave alone, and what doesn't even matter...

Any advice would be incredibly helpful!

I will try and post later today what I have been slowly working on. Real work has me super busy so things have slowed down for the holidays. I have been hyper focused on an IntoRealism Ultra v5.0 prompt set with as much data as I can get.

Here's an example output:

 

[cxxxgaming]

hold up how'd you cut out the background? i get a big ass white background and nothing else for whatever reason

hey i think i can help you with that. Just change the color code in SDClient_Ren.py to #000000

backgroundcolor = "#000000"; # leaving black if we don't try to remove it, might be easier to replace for stable diffusion than a solid purple?
if persistent.sd_removebackground_enabled:
backgroundcolor = "#ff00ff"; <---- this one here needs to be changed

 

[The Huntsman]

I will try and post later today what I have been slowly working on. Real work has me super busy so things have slowed down for the holidays. I have been hyper focused on an IntoRealism Ultra v5.0 prompt set with as much data as I can get.

Here's an example output:

View attachment 5505680

Thanks for the quick reply, and that's astonishing!

Here, if it might save you some trouble, I'll list the questions I have - and I'll also keep looking for these answers in other places and doing other testing, so you don't have to feel responsible for answering them all or anything, lol.

First, I've somewhat figured out how to make my prompts auto-populate based on a lot of factors (I pretty much copied that list of position prompts you posted last month, for example), but so far I only know how to EDIT the things that are already called up for prompts in that SD_PromptBuilder_ren.py, and I don't know how to ADD things to it. For example, I really want to add a section where it checks if they're wearing any neckwear, like a choker or a necklace, and then add that. Additionally, I want to make it check their arousal, or whether they're in a trance, to make their expressions change appropriately. How would I add these things? Can I do it with just that file or do I need to go deeper?

Also, is there a list somewhere of all the source-code terms for various items, like clothing, positions, stats, etc.?

Lastly, and this is specifically for you, that pic looks fantastic, and so far I haven't had a ton of luck with lifelike models, just some high quality artistic ones, but the goal is to get to the lifelike ones. Do you find they generate properly with consistence? Can can you tell characters apart easily like with a LoRA or something, so that, say, Stephanie always has Stephanie's face - or at least close to it? Have you had success?

Any help you have time to give is greatly appreciated, so thank you again!

 

[Bobincob]

I will try and post later today what I have been slowly working on. Real work has me super busy so things have slowed down for the holidays. I have been hyper focused on an IntoRealism Ultra v5.0 prompt set with as much data as I can get.

Here's an example output:

View attachment 5505680

Bro, I'd love to see the generation data for that image or even what you've got set up in your JSON prompt file, because I've been trying to use the IntoRealism Ultra v6.0, and I don't know if there's just THAT much difference between version 5.0 and 6.0, but my images have been coming out like shit lol.

 

[The Huntsman]

Bro, I'd love to see the generation data for that image or even what you've got set up in your JSON prompt file, because I've been trying to use the IntoRealism Ultra v6.0, and I don't know if there's just THAT much difference between version 5.0 and 6.0, but my images have been coming out like shit lol.

I just switched to CyberRealistic Pony v14.1 to try to get some like that and IMMEDIATELY had incredible results. Granted, I was massaging the prompt pretty significantly. All of mine start with this in the positive:

ultra photo-realistic, best quality, ultra-detailed, realistic, 1girl, solo, pov, human,

And this in the negative:

(worst quality:1.2), (low quality:1.2), lowres, bad anatomy, bad hands, missing fingers, watermark, loli, child, bokeh, symbol, logo

Then, I've got the prompt builder pretty heavily customized already, but I started a new game to test it and got this for Jennifer right out of the gate (after I changed her hair color, of course, lol).

 

[NaughtyAnon]

Thanks for the quick reply, and that's astonishing!

Here, if it might save you some trouble, I'll list the questions I have - and I'll also keep looking for these answers in other places and doing other testing, so you don't have to feel responsible for answering them all or anything, lol.

First, I've somewhat figured out how to make my prompts auto-populate based on a lot of factors (I pretty much copied that list of position prompts you posted last month, for example), but so far I only know how to EDIT the things that are already called up for prompts in that SD_PromptBuilder_ren.py, and I don't know how to ADD things to it. For example, I really want to add a section where it checks if they're wearing any neckwear, like a choker or a necklace, and then add that. Additionally, I want to make it check their arousal, or whether they're in a trance, to make their expressions change appropriately. How would I add these things? Can I do it with just that file or do I need to go deeper?

Also, is there a list somewhere of all the source-code terms for various items, like clothing, positions, stats, etc.?

Lastly, and this is specifically for you, that pic looks fantastic, and so far I haven't had a ton of luck with lifelike models, just some high quality artistic ones, but the goal is to get to the lifelike ones. Do you find they generate properly with consistence? Can can you tell characters apart easily like with a LoRA or something, so that, say, Stephanie always has Stephanie's face - or at least close to it? Have you had success?

Any help you have time to give is greatly appreciated, so thank you again!

Work done and bourbon in hand. Lets get this part started.

I'll start with the core information. I am using the mod package generously provided by gimb here and for the game I use SD Forge set to the XL UI to adjust memory use. So far this has allowed me to focus on using the prompt files to do all the heavy lifting and this where I have modded the most. For ease of use I added the json file I have been adjusting and currently run the game with.

The prompts are mostly driven by the help of two textural inversion files (

You must be registered to see the links

and

You must be registered to see the links

) and one LoRA (

You must be registered to see the links

). I have played around with position loras for specific ones that have been giving me trouble but the mutations have not been pretty. There are a lot of outfit substitution tags and color adjustments in the version I run and I am sure it can also include the chokers you're looking for. To find the clothing terms you have to dig through all the source files and hunt them out. I did it for positions and haven't had the energy to dive back in for more.

As far as generating characters so long as the seed doesn't change for the person they have always looked the same even when changing hair style and color. I think between IntoRealism Ultra and the few extra positives and negatives everything has been pretty darn stable. Strangely enough v5 has been working better than v6 so that's something I may dig into later after the insanity of the holidays.

The Huntsman I love CyberRealistic too this is just one Ive had really good luck with lol.



View attachment IntoRealismUltra5_Forge.7z

 

[The Huntsman]

It's hard to argue with bourbon and good results! I keep forgetting that the seed is probably key to maintaining their faces and I just end up resetting to a new seed when I get creepy mutations... But I might need to break that habit.

Okay, knowing that that's how it works then I guess I just need to go hunting for any specific clothing items or acts that I might want to focus on. I was going to specifically ask about adding prompts based on location, and it looks like you have them in here, so I'll dissect that a bit and see if I can apply it to my own setup. Thank you for sharing that, by the way!I also like that you added times of day, I wouldn't have thought of that, but it's cool!

What does this modded-mod add that isn't in the original mod? Is it something that I wouldn't already be able to have done myself?

 

[waaazzzaaa]

Attached is the V0.01 alpha release of a mod I'm working up from AnotherMike's narration text version. It currently only works properly with the

You must be registered to see the links

but I'll be adding more over time. It's meant to better fit generated GenAI images to what's happening in the game. Eventually I'll train a checkpoint optimised to LR2 image requirements and post a link in this forum.

For CyberPony user, I still haven't found a modded mod that works better than that the one from Crudi. It also works with almost any checkpoint that uses (weights:1.5) and ((Priority)) prompt style. And has some speciffic narrated scenes which are great.
I've made a lot of additions to the prompt builder though. I'm trying to make it so the generated image includes as much as possible from the VT tracker and game UI.
Arousal past 50% = moist down under
Arousal past 110 = wet
doses of seed 1 = drip, 2 = ooze, 3 = overflow
baby bump grows as it get closer to delivery etc...

Spoiler: Prompt addditions

You don't have permission to view the spoiler content. Log in or register now.

All this make the game much more immersive as the effects of flirting and doses or trances linger but that also means A LOT MORE generted images per intraction. So if your generations take more than 15sec it might not be worth the waits.

I suggest using the LR2_AIO_Script.bat (just google that) it will install LR2 dev version and update dev version every time you run it after. Seems to be a minor script update about every 3 days or so... they finally fixed infinite erection bug where you never had to change position... you can still continue past zero energy though.

Also, I highly recommend using StabilityMatrix to install and manage SD. It basically automates installing a1111 or forge or comfy and keeps all checkpoints centralized. Forge classic seems to be the best balance for me on 32gb ram, 8GB vram. CyberPony DPM++2M, Karras, at 832x1216 gens are about 10 seconds each. StabilityMatrix also has a built in checkpoint browser and you can directly update or download from hugging face and CivitAI - just gotta add your api keys. Make sure to check CivitAI for each checkpoint's suggested settings. Look for "Show More" in the info or somtimes in the "About This Version"


If you have 8gb vram like me, and use 6gb checkpoints, I suggest adding " --always-high-vram" to Forge, set COMMANDLINE_ARGS=. It will keep the checkpoint in vram ALWAYS, so no more loading from ram... but if you use loras it could backfire and keep that checkpoint in vram and your lora would end up overflowing vram and running in regular ram. Ram is rouglhy 10x slower than vram.

All that being said, ComfyUI and Z-Image seem to be the future. So it so might be worth waiting a bit if you are new to SD local AI and have 8gb vram or lower.

** anybody know how to get Forge Classic - Neo, to run Z-image? They added support for it but I just get "Error: You do not have Qwen3 state dict!"

*** (A cyberpony 15.0 will be public in 11 days)

 

[The Huntsman]

For CyberPony user, I still haven't found a modded mod that works better than that the one from Crudi. It also works with almost any checkpoint that uses (weights:1.5) and ((Priority)) prompt style. And has some speciffic narrated scenes which are great.
I've made a lot of additions to the prompt builder though. I'm trying to make it so the generated image includes as much as possible from the VT tracker and game UI.
Arousal past 50% = moist down under
Arousal past 110 = wet
doses of seed 1 = drip, 2 = ooze, 3 = overflow
baby bump grows as it get closer to delivery etc...

Spoiler: Prompt addditions

You don't have permission to view the spoiler content. Log in or register now.

All this make the game much more immersive as the effects of flirting and doses or trances linger but that also means A LOT MORE generted images per intraction. So if your generations take more than 15sec it might not be worth the waits.

I suggest using the LR2_AIO_Script.bat (just google that) it will install LR2 dev version and update dev version every time you run it after. Seems to be a minor script update about every 3 days or so... they finally fixed infinite erection bug where you never had to change position... you can still continue past zero energy though.

Also, I highly recommend using StabilityMatrix to install and manage SD. It basically automates installing a1111 or forge or comfy and keeps all checkpoints centralized. Forge classic seems to be the best balance for me on 32gb ram, 8GB vram. CyberPony DPM++2M, Karras, at 832x1216 gens are about 10 seconds each. StabilityMatrix also has a built in checkpoint browser and you can directly update or download from hugging face and CivitAI - just gotta add your api keys. Make sure to check CivitAI for each checkpoint's suggested settings. Look for "Show More" in the info or somtimes in the "About This Version"


If you have 8gb vram like me, and use 6gb checkpoints, I suggest adding " --always-high-vram" to Forge, set COMMANDLINE_ARGS=. It will keep the checkpoint in vram ALWAYS, so no more loading from ram... but if you use loras it could backfire and keep that checkpoint in vram and your lora would end up overflowing vram and running in regular ram. Ram is rouglhy 10x slower than vram.

All that being said, ComfyUI and Z-Image seem to be the future. So it so might be worth waiting a bit if you are new to SD local AI and have 8gb vram or lower.

** anybody know how to get Forge Classic - Neo, to run Z-image? They added support for it but I just get "Error: You do not have Qwen3 state dict!"

*** (A cyberpony 15.0 will be public in 11 days)

This was all helpful AF, just so you know, lol.

You guys are awesome!

 

[waaazzzaaa]

found another variable: "is_highly_fertile" (that honeybee icon)

done almost no testing with this as I'm trying to figure out how to include Bio, slut, breed collars.

Spoiler: Highly Fertile

You don't have permission to view the spoiler content. Log in or register now.

anyone have more than just slave collar working in their prompts?

Spoiler: Collar Accessories

You don't have permission to view the spoiler content. Log in or register now.

 

[NaughtyAnon]

It's hard to argue with bourbon and good results! I keep forgetting that the seed is probably key to maintaining their faces and I just end up resetting to a new seed when I get creepy mutations... But I might need to break that habit.

Okay, knowing that that's how it works then I guess I just need to go hunting for any specific clothing items or acts that I might want to focus on. I was going to specifically ask about adding prompts based on location, and it looks like you have them in here, so I'll dissect that a bit and see if I can apply it to my own setup. Thank you for sharing that, by the way!I also like that you added times of day, I wouldn't have thought of that, but it's cool!

What does this modded-mod add that isn't in the original mod? Is it something that I wouldn't already be able to have done myself?

So to cover the quick answers. The modded-mod mostly helps refine the image and replace the background with something that reflects the area you are in. I also noticed it just naturally did great with collars and I didnt have to do anything special to get them to look correct. And now on to the fun stuff....

Welp fixed the issues with IntoRealism Ultra v6 and it has fixed a few of the previous issues. Just needed to ramp up the sampling steps and things have come out a bit nice. Here's a few of the new generations.



On a side note I have gotten the mod by

You must be registered to see the links

to work with ComfyUI. The main issues I have is it only works with pyTorch and I use SageTorch in Comfy and any time it tries to use the openpose references it trips, impales itself on a spike, and loops 0-1% over and over. So any help there is appreciated. But Comfy is amazing especially with the new Z-Image workflows.

ComfyUI Z-Image tests:

 

[gimb]

What does this modded-mod add that isn't in the original mod? Is it something that I wouldn't already be able to have done myself?

Maybe i can help with that from what i posted a month ago on discord

1. a few extra prompts that are not (yet) supported by the json like location, time of day, height and weight
2. a bug with img2img reading of the Json
3. support for an forge extension (integrated latent modifier) that helps with img2img generation
4: changing the background of img2img base file to green
5. the option to add an universal weight to most prompts so i.e. all clothing become (clothing:1.3)
6. Fixed a thing in the ethnicity prompt where it it gives "{value} skin" as transfer value so you cant make the json read for example "(dark skin:1.2)" just "(dark:1.2) skin.

All Jsons should be compatible beteween maswoo's and my version it's just that those additional things only activate in my version.

Currently there is a version 6 but i just mostly changes a few age height weight tresholds

 

[The Huntsman]

You guys are clearly further into this than I am and I’m excited to be learning this because I haven’t worked much with AI image generation before the last couple weeks.

NaughtyAnon tonight I’m gonna probably swap to Into Realism Ultra V6 so that I can maintain some uniformity with you and make troubleshooting easier. gimb I’ll also get your modded mod set up for the same reason, especially because you’re already working on things I’ve been wanting to do, lol.

I spent about an hour last night trying to get locations to auto-prompt and made no real headway, so this seems like a smarter route.

EDIT: I worked with gimb on it a bit this afternoon and it's going great!

 

[NaughtyAnon]

Ran out of time this morning before having to work the normal job but, I am jumping into testing Forge Neo to see if I can get Z-Image to work with the mod and forge prompting I've already done. Will update tonight after work.

Update: I got it to work. A combination of setting things within Forge Neo and a few tweaks in the prompt set makes it work just like the normal forge setup I have running. Output it a bit slower, 30 seconds vs 10 seconds, but I think there is potential.

 

[tpogs]

This work with ComfyUI? Don't really want to set up another workflow in Forge for this.

 

[NaughtyAnon]

This work with ComfyUI? Don't really want to set up another workflow in Forge for this.

I'm not sure if this question is directed at my post but I'll give some information and an update.

So the Z-Image Turbo sets out now do work with ComfyUI workflows and can produce amazing results. I unfortunately seem to have broken the mod I was running that talked to ComfyUI so thats when I gave Forge Neo a try.

Here is what I have discovered so far.

1. This process is in VERY early testing as we all figure out the new workflows.
2. It DOES NOT do well with sexual prompts (positions, male anatomy, etc.)
3. There is a big change in the prompting style so everything that was (prompt:1.5) needed to be edited.
4. When it goes bad.... woooo it goes really bad.
5. I have gotten it down to 7-9 seconds per generation.

Overall I think shows great potential but if you just want to enjoy the GenAI in game there are better options currently.