Links and Resources
https://lorelai.ai - My company website
https://www.youtube.com/@misogenius
https://twitter.com/lorelaiai - follow me on twitter
Loras and Checkpoints I Recommend
DucHaiten-AIart-SDXL - v2.0 | Stable Diffusion Checkpoint | Civitai
Shambhala.XL - Artistic.Soul | Stable Diffusion Checkpoint | Civitai
SDXL-Anime-Deephentai-mix - v1.32 | Stable Diffusion Checkpoint | Civitai
hyper bottm heavy SDXL - v1.0 | Stable Diffusion LoRA | Civitai
breasts adjust_XL(胸部大小调节器) - v2.0 | Stable Diffusion LoRA | Civitai
Bombshell XL - big busty boobs - v2.0 | Stable Diffusion LoRA | Civitai
Clothing Slider SDXL - v1.0 | Stable Diffusion LoRA | Civitai
Disclaimer for Using My Lora Files
As the creator of the Stable Diffusion Lora Models, it is important to note that these models have been specifically designed for my use and have not been optimized for public use. While I am sharing them in the hope that they may benefit others, it is important to understand that they may need to be more efficient and optimized than models specifically trained for public use.
I will only make major optimizations to these models if they can be easily corrected with feedback. It is highly recommended that you follow the instructions for the best results when using these models.
Requirements and Recommended Settings
Requirements:
To run Stable Diffusion Lora Models, it is necessary to have a computer capable of reasonably running SDXL iterations. This typically requires a computer with at least an NVIDIA Graphics Card with a minimum of 12GB of VRAM. It is recommended, when possible, to use a more powerful graphics card for quicker image generations, such as the Gigabyte GeForce RTX 4090 Overclocked.
Recommended Settings:
It is recommended to use Automatic1111 in combination with the SDXL Style Selector Extension for image generations. Image generations for mobile portrait should be done at 768 by 1344, while landscape images should be at 1344 by 768. Trainings were done 1024 by 1024, as this is optimal for SDXL. It is advised not to generate images smaller than what I have stated, as it will result in degradation in many cases. As for the sampling method, DPM++ 2M Karras or Euler are usually sufficient, with 20 steps and 7 CFG. It is also recommended to upscale images independently outside of Automatic1111 and avoid Hires Fix. In the SDXL Style selector, options such as Hyperrealism, Fantasy Art, or Anime are popular. Additionally, the checkpoints DucHaiten, Shambhala, and SDXLAnimeDeepHentai work well with most of my loras at a weight of .8.
Overview of my Lora Files
Most of my trained models in the Stable Diffusion Lora are female characters, with a few exceptions of male characters. The models predominantly consist of anime characters, significantly focusing on the Dragon Ball series. Moving forward, I will prioritize creating models based on demand, particularly with a current emphasis on female characters. Also be aware that 100 percent of my models and work is with SDXL.
As I develop new models, I will closely consider the popularity of certain characters and series to ensure they align with users' interests. This will allow me to tailor my models to the audience's preferences and provide the best possible experience for those interacting with the Stable Diffusion lora files.
How to Approach Prompting with My Models
To approach prompting with Stable Diffusion Lora Models, start by choosing a checkpoint and setting the weight to .8. You can also select a style from the SDXL Style Selector or stick to the base. When generating an image, lower the weight if it looks too much like the anime or the model. If it doesn't resemble the model enough, raise the weight. For modifying a woman's body, consider using resources like BottomHeavy, BreastsSlider, or BombShell. This process is meant for experimentation and fun, so please feel free to play around with it. These instructions are simply a starting point to get you going with the model. Remember that the recommended checkpoints are just suggestions, and you should feel free to explore other options that suit your preferences.
Getting Extra Help
I'm here to help you with extra help with the Stable Diffusion Lora Models. Just to let you know, I am here to help with prompting and not training. Please feel free to join my Discord (insert link here) where I will be available to answer any questions or give you additional guidance within reason and availability. I aim to ensure you thoroughly understand the Stable Diffusion Lora Models and can effectively apply them in your work. By joining my Discord, you will have access to a community of individuals who can offer support and insights.