Recommended resolution: Short side 1152+, long side 1920+. Landscape focuses on the scene, portrait focuses on the character.. Hirs repair can be opened at 1.5 to 2 (optional), no trigger words.

The base model used for training isFlux.1-Dev,

How This Model Is Trained

• This model is trained with kohya-ss/sd-scripts, the images are generated with lllyasviel/webui-forge

• The auto-training framework is maintained by DeepGHS Team, fork by me.

• The step we auto-selected is 200 ep to balance the fidelity and controllability of the model.

Why Not Just Using The Better-Selected Images

Our model's entire process, from data crawling, training, to generating preview images and publishing, is 100% automated without any human intervention. It's an interesting experiment conducted by our team, and for this purpose, we have developed a complete set of software infrastructure, including data filtering, automatic training, and automated publishing. Therefore, if possible, we would appreciate more feedback or suggestions as they are highly valuable to us.

Why We Can't Accurately Generate the Desired Art Style

Our current training data is sourced from a variety of image websites. Predicting the specific art styles present in official images within a fully automated pipeline presents a significant challenge. As a result, the generation of art styles relies on clustering based on labels from the training dataset, striving for the best possible reproduction. We will continue to tackle this issue and seek optimization, but it remains a challenge that cannot be entirely resolved. The accuracy of art style reproduction is also unlikely to match the level achieved by manually trained models.

For the following groups, it is not recommended to use this model and we express regret:

1. Individuals who cannot tolerate any deviations from the original art style, even in the slightest detail.

2. Individuals who are facing application scenarios with high demands for accuracy in art style reproduction.

3. Individuals who cannot accept the potential randomness in AI-generated images based on the Stable Diffusion algorithm.

4. Individuals who are not comfortable with the fully automated process of training art style models using LoRA, or those who believe that art style training must be done purely through manual operations to maintain the integrity of the original artistic vision.

5. Individuals who finds the generated image content offensive to their values.

Why Do We Use Auto-Training Framework

I just need to put the images there, and the rest of the steps are fully automatic, requiring no supervision. I can go play games or hang out, completely without the need for any tedious data processing and labeling work, and the speed of data processing and training is very fast. This method of model training is a complete enjoyment for me.