Have been practicing with no success on large models. After more than two months of twists and turns, I finally have this current version of a fairly decent large model.

The main purpose of refining this model is for those imaginative and sexy themes that have always interested me yet never found a base model that suited my taste. I've practiced a lot, but was lacking a particularly powerful base model. It was with this simple thought in mind that I began trying to refine the base model.

The dataset for this base model consists of CG artworks that we like. Because CG artworks are currently the only way to visually express the themes of gods, demons, ghosts, and monsters sensually, 3D animations occasionally touch upon similar effects, but are still rare. As a result, the process started with CG art as a base and gradually incorporated some 3D effects. Due to my limited technical knowledge, I could only perform simple adjustments and blends. Despite repeated tests with my own lora and prompts, I always found myself in a dilemma between realism and fantasy, tradition and current trends, even a sense of futurism. Sometimes a group of lora and prompt effects would improve, only for me to realize later that the previous ones had lost their accuracy. Hence, I deeply admire those creators of large models; indeed, a good large model is the result of assiduous refinement.

Since I have embarked on the path of refining the base model, I will continue earnestly. There should be updated versions in the future, at the very least with different inclinations.

Hats off to the many talented alchemists of large models!

Based on my own evaluations, this model basically matches my own lora; I will slowly share some of the effects that I have used in the feedback area. I also welcome partners to showcase their skills, and remember to share good results in the feedback area to impress me~

Start with these few:

National Style Yao Dan: https://www.liblibai.com/modelinfo/7c409ec65000330eb6f82ae3eef7d0d5?mine=1

Dark Giant Dan: https://www.liblib.ai/modelinfo/d1a7713537714967bac8452e1356b4d2

Mountain and Sea Fairy Dan: https://www.liblib.ai/modelinfo/51ab54096f1a4f6c9a00924710c882e5

Hell Dark Dan: https://www.liblib.art/modelinfo/2b9bd5d0366b4c58b845a30a64feddbc

National Style Endless Mountains and Waters v2.1: https://www.liblib.art/modelinfo/cd097baa3ab64538a8a0b9667115f654

【Usage Guide】

1- Recommended CFG: 3-11.

This model is quite sensitive to CFG values. The higher the value, the more realistic the output, and it pays more attention to the logical narrative of the scene, which is understandable as CFG controls the semantic weight of the prompt. It is not recommended to exceed 11, as it may lead to collapse phenomena such as the appearance of some unknown particles. The optimal range is usually between 5-9, but sometimes exceeding normal values can have miraculous effects, for example, reducing to 1 may produce a wonderful artistic style and brushstrokes. It will be closer to traditional Chinese painting style, with very light color layers, which makes it seem quite flat, but this is a characteristic of traditional Chinese aesthetics, delicate and gentle.

2- Recommended Samplers: DDIM, Euler a

DDIM is good for imaginative works suitable for letting the imagination run wild; on the other hand, Euler a is more logical, thus more normal in composition and element construction. Both are well-established and reliable samplers.

3- Strengthening Word: "zg_style". It is advisable to use "zg_style" as a trigger word at the beginning of the prompt. In principle, this model does not require a trigger word, but when you need a more Chinese style atmosphere, you can strengthen it through this word. You can also add "ghost_cn" to enhance the atmosphere, not necessarily a dark atmosphere but summoning elements like mist and clouds, which helps to create an atmospheric rendering. The specific effect depends on the theme and other parameters.

Enjoy your creativity!

(Note: The cover images are the effects of using multiple style loras! The original images with parameters and direct output effects will be gradually posted in the feedback area~)

Thank you for reading! Let's exchange feedback with more images~

Disclaimer:

In principle, this model is for educational purposes only, please do not use it for commercial purposes.All content generated based on this model is the user's independent creative action after judgment, please comply with relevant laws and regulations.Any intellectual property issues arising from using this model to generate content will be your own responsibility.Without the author's written consent, you may not integrate this model service into any third-party software or service, nor disclose the relationship between the generated content and this service to any third party in any direct or indirect, explicit or implicit manner. Any losses incurred by the author due to your violation of this agreement will give the author the right to recover from you.