AndroFlux is a LoRA model made for FLUX.1 dev by Black Forest Labs, because there’s a glaring lack of accurate male anatomy in AI models. I’m trying to fix that by making sure AI can depict male features more accurately, balancing out the overwhelming female-centric content you see everywhere.
Community and Support
Join The Bulge Discord at https://thebulge.xyz if you’re into gay AI art—it’s a great place to share and learn. If you want to support the project, whether with feedback or donations, it’s much appreciated.
Current Status
The model isn’t perfect, so don’t expect flawless results.
Purpose
AndroFlux is all about making sure AI models can represent male anatomy—especially penises—accurately. With that said, it also has a general photographic aesthetic and focuses on twinks so it is still useful for other SFW purposes. It’s a step towards more balanced AI-generated art.
Technical Details
AndroFlux was built and tested primarily on FLUX.1 dev, but it should also be compatible with the Schnell base models, including the quantized fp8 version. These are distilled models, making training tricky and affecting the LoRA’s performance, requiring a lot of optimization.
Training Process
Resolution Journey: I started training at 1024px, but the model wasn’t converging, so I dropped to 512px. The quality wasn’t great, so I ended up at 768px. This resolution offers a good balance between detail and training time.
Captioning: Captioning was crucial in training. Contributions came from the AndroAtlas dataset, part of the AndroFlow project. Ephemeron01 from The Bulge Discord assisted with captions using models like llava-hf/llava-v1.6-mistral-7b-hf and a custom-trained PaliGemma VLM. Specific prompts like "man, penis" and the activator "AndroFlux" were also used in later versions.
Acknowledgments: Huge thanks to bghira and Terminus Research Group for their support, including providing an A100 for some of the training runs.
Usage Tips
For best results, use prompts that specifically mention male frontal nudity, such as "penis."
Recommended settings:
Resolution: 1024x1024 or 832x1216
Flux Guidance: 3.5
LoRA Weight: Start at 1.0, reduce to 0.5 if artifacts appear
CFG: 2.0 (1.0 will be faster though)
Suggested Prompts
A nude man in his {late teens | early twenties | early thirties}
His penis is {erect | flaccid}
His penis is {large | small | average}-sized
Version History
See "About this version" on the right for version-specific updates.
Description
This is the initial test version of the v5 training run on RTX A6000 Ada
FAQ
Comments (16)
Please share the training parameters (LR, batch size, grad. accumulation, optimizer) if it is not a secret.
FLUX is so slow and hard to learn unfamiliar things and styles (compared to SDXL) that it can cost a fortune to experiment blindly...
For this training run I decided to go with Prodigy so I wouldn't have to deal with LR and all bare minimum batch sizes to make sure it would actually run without OOM-ing. I followed this config file that was discussed in the SimpleTuner github (it requires you to request access first): https://huggingface.co/mikaelh/flux-sanna-marin-lora-v0.2-fp8/blob/main/config.env
If I change any of the settings, I'll be sure to post them in the description. For now I'm focused on getting past this training hurdle where it turns everything to mush and actually getting a usable LoRA (and getting those updates out). Thanks for riding along on this adventure ~
Why make it a LORA and not a DORA? It is a lot more accurate and works better for FLUX.
How much better is it? Are your DoRAs working well?
@markury it's not just better in terms of accuracy, it's better to train Doras for flux in general, since they don't suffer from the problems you are having with Loras training. My comment is intended to make your life easier btw. There is a discussion about training here:
I'll give it a try for the next version then. Thanks!
It looks like it has no effect on flux1-dev.sft What's going on?
Make sure you load the LoRA in comfyui with the LoraLoaderModelOnly node at around 0.5 weight. Connecting the CLIP will ruin your images and other LoRA loaders might not work with Flux yet.
thanks for working this! But I hope it's not gonna turn every man into a twink!? 😭😆
fabtastic! Can't belive a penis lora is one of the first flux loras. But how do we use it on comfyui? Do you have any worflows or tips? do we just use any lora loader?
Answering myself, I just tested the Power Lora loader and it works. Although the LoRA itself has a long road ahead. The image quality suffers, the text ability also suffers, and the penis is nowhere near a penis. But it's expected as a first version, Penises are historically a super hard concept to bring back to models.
You should use the LoraLoaderModelOnly node that is included in comfyui and no other nodes. Connecting the CLIP will ruing your image because I didn't train the text encoders. It's still in it's early stages right now so I would say max strength 0.5 to avoid getting random mush images. As long as you don't touch the text encoders, text comprehension does not seem to suffer (as seen in the demo images). Here's the workflow I'm using for basic txt2img: https://civitai.com/models/618997/simpleadvanced-flux1-comfyui-workflows
@markury after some more testing, the quality only suffers at higher weight. So around 0.5-0.71 it is still good. Also the text ability is better and don't suffer if we increase the guidance. Of course the penis is nowhere near good but it's a good first step. Thanks! Hope you keep developing it.
Also as another person mentioned, Doras are much better ! please research this new technology !
I am waiting on Comfy to commit the DoRA changes. Right now if I released a DoRA you could only use it through diffusers. But it's on my radar.
Great effort! 🍆♥️