This is my first ever attempt at training a Stable Diffusion XL model, and while I'm excited to share it with the world, I want to be transparent about its current state. As a first iteration, you might encounter some limitations and imperfections.
Known Issues:
Limited Training Data: Due to the intensive nature of training SDXL models, access to vast amounts of data is crucial. My initial training dataset might be smaller than ideal, potentially leading to issues like inconsistencies or limitations in generated images.
Fine-tuning Needed: Further fine-tuning with specific prompts and styles is likely necessary to achieve optimal results. The model can be further improved upon by incorporating additional training data and refining the conditioning process.
Beta Stage: Consider this model a work in progress. I'm actively working on resolving identified issues and improving the overall performance.
Recommended parameters:
I don't use refiner. I just don't like it. But feel free to use it if you want.
Sampling: DPM++ 2M Karras | work pretty good for me
Steps: 20 | 40
CFG: Under 5/10
Upscaler : 4x-UltraSharp
