💪​🍆​ GodMaker

Complete LoRA for hyper muscles and penis

Introduction

This LoRA purpose is to add (a lot) of size to your SDXL image generation. The base model I used for training is Photonic Fusion SDXL, as it supports natively NSFW topics and elements (such as sex and penises). I strongly recommend to use this model as base for your generations to ensure best results.

Trigger words

There are trigger words individually for muscles and for penises.

Because of the broad range of the concept of "hyper", and for maximum versatility, I chose to categorize the strength of hyperization : big, hyper, and colossal.

Here's a breakdown on what it means :​

💪​ Muscles

--musclesbig ➡️​ a very muscular man, but still somewhat realistically muscular. If it is, it's definitely the world's top physique.

--muscleshyper ➡️​ superhuman with massive arms, pecs and legs, emphasizing on very broad and round shoulders, and visible traps. Nobody could be that muscular, right ?

--musclescolossal ➡️​ forget realism. We're talking about a man that is wider than he is tall, almost smothered in his own muscles. Every muscle is bulging and could very well have a mind on their own. Almost ridiculously too big (but we love it).

🍆​ Penis

NB: Penis size as discussed below is relative to the body (meaning a "big" penis will not have the same absolute size on a "big" muscle body and a "colossal" body)

--penisbig ➡️ size ranging from "definitely above average" and "huge". Still somewhat realist, it allows a man to have a penis as big as his forearm.

--penishyper ➡️ grows one man's penis to become almost a third leg, being as big as a whole arm. It's still holdable (both hands are just falling short), and is long enough to reach under the knees most of the time. The testicles are as big as tennis balls.

--peniscolossal ➡️ create an over sized penis as big as its owner (or even bigger). You could hold it close against your chest and it would be taller than you. Given the weight, the balls are likely touching the floor.

Generation tips

• Please refer to the metadata of this model's images collection to get a good example of what's possible and with which parameters. Your imagination do the rest.

• Some generic keywords enhance greatly the hyper aspects : thick arms, thick legs, large pectorals, thick penis, large penis...

• Keep the CFG low, around 2.5 to 5. Otherwise the images get easily that overexposed or deep-fried look.

• Model strength works well at 1. You can lower it a bit if the thickness is too much for you.

• Don't forget the two dashes "--" in front of this LoRA's keywords, they are part of it.

More about the LoRA and its creator

This LoRA was trained on 325 unique hand-picked high quality images, manually sorted and resized, featuring each size in similar proportions. Almost all images were realistic-style pictures, with a few artworks sprinkled in for the harder concepts like the colossal sizes.

Training captions were generated automatically by WD14 Tagger.

I trained this LoRA locally on my computer using Kohya-ss !

I recently discovered AI image generation, and this is my first "real" model training project that I am confident I can share with you guys !

Please share your feedback and suggestions so I can improve ! and maybe push out an even better v2.0 ? 💪​

Known issues and limits

Please share any difficulty you encounter using this model. I personally will continue to generate a wide range of scenarios to test its limits in the hope of creating a better version in the future.

Here some weaknesses I encountered this far :

• --musclescolossal is too mild (on model strength 1.0). It is probably under trained. To ensure real colossal result, add weight to the keyword in your prompt like this : (--musclescolossal:1.3). Works best under 1.5. Will be fixed in next version.

• There is a strong correlation between muscle size and penis size. Usually, more muscle means more penis. It is incredibly hard to generate edge cases (twink with huge penis or muscle god with regular penis). You can mitigate this by putting muscle/penis-related keywords in the negative prompt. I don't think this can easily be fixed with my current training data. Probably won't focus too much on fixing.

• It seems harder to generate other ethnicities than caucasian, mainly black men. This is probably due to the training data biased towards white men, and the few images depicting black men being miscaptionned. Training data will be adjusted to try to fix.

Thank you for using my work !

🍌​ Ahsolan, muscle enthusiast