⚠️ An Important Personal Note & The Reality of Future Models
I want to be completely transparent with all of you: producing future models and updates is going to be extremely difficult for me moving forward.
I am currently going through a very tough period in my personal life, which has taken a heavy toll on me emotionally, but most critically, financially.
To give you an idea of my workflow:
I train all my models on cloud instances (like RunPod) using high-end hardware such as an RTX 5090 or an RTX 6000 Pro, depending on the size and demands of the base model.
Locally, my personal rig only has an RTX 5060 Ti (16GB). While this card is great for testing and generating without issues, full-scale training on it takes an eternity, and I refuse to sacrifice quality by relying on heavily quantized training methods.
Because cloud compute costs real money that I currently cannot afford, my workflow is heavily bottle-necked.
With that said, I will NEVER put my models behind a paywall.
I firmly believe in the open-source ethos of AI. Just as we all want base models to be accessible and free, I believe LoRAs, workflows, and community tools should follow the exact same philosophy. Art and tools should be accessible to everyone.
-Thx for your time reading this.
I’ve had a blast experimenting and generating with this new base model (Moss). It is incredibly versatile and consistently produces genuinely distinct results—it truly brings back that pure excitement from the early days of Stable Diffusion!
I trained this LoRA using a dataset composed almost entirely of real-world photography. My main goal was to achieve natural skin textures, genuine imperfections, and authentic lighting, completely moving away from the over-polished, "plastic" look typical of purely synthetic datasets. While it may not be mathematically "flawless," it excels at delivering realistic, tangible results.
✨ Key Features & Usage Tips
No Hard Trigger Word Needed: You don't necessarily need a strict trigger word to see the effect. However, if you want to push certain proportions or features to their absolute limit, simply adding emphasis keywords like huge or big will do the trick.
Balanced Body Types: Models trained on real-world photo datasets often tend to bias heavily toward curvier or thicker builds. I did my best to balance this out. If you prefer a leaner physique, simply adding skinny or slender to your prompt should correct the body type smoothly without fighting the model.
🧡 How You Can Help
Because I want to keep everything 100% free for everyone, I have to rely on community solidarity.
If you appreciate my work and have the financial flexibility to do so, even a small tip of a couple of dollars makes a massive, tangible difference (I’m based in Europe, so it converts straight into € to help directly cover cloud GPU hours and keep the lights on).
Please only donate if you can comfortably afford to do so.
☕ Support my work on Ko-fi:
Beyond donations, your kind comments, reviews, and the images you post are the real fuel that keeps me motivated to push forward during dark times. Thank you from the bottom of my heart for all your support!