TY Han SDXL / 天韵古风大模型

https://civitai.com/models/133228

Trigger Words: hanfu, chinese clothes

Chinese clothing culture XL Version 1.0 is officially released, which solves the watermark problem and makes the pictures more realistic. 1.0版本正式发布，解决了水印问题，出图更加真实

炼丹交流群：33529439，进群密码：civitai

1.0 detail: Suggestions for inference: Adding "hanfu" and "chinese clothes" can better present the clothing style , Since a large number of artistic photos are used for training, adding ((blurry background)) to the reverse prompt words can make the background more realistic

cfg: For example, around 3 can better increase the sense of realism. 5.5 is a more suitable cfg.

sampler: I use dpmpp_2s_a which is pretty good.

I'm not very good at inference pictures. I hope you can return more pictures and give some good effects.

1.0细节: 推理建议：增加"hanfu" 或者 "chinese clothes" 可以增加更多的风格化，由于训练使用了大量的艺术照片，所以反向提示词中加入((blurry background))可以让背景更加真实

cfg: 低真实度 比如3左右能够更好的增加真实感，5.5是比较合适的cfg

sampler: 我用dpmpp_2s_a 挺不错的

我对于推图不太在行，希望大家可以多返图，给出一些好的效果

If this model has helped you, please give me a ⭐️⭐️⭐️⭐️⭐️ review add a ❤️ to receive future updates.Consider supporting me on Patreon

Hanfu, as an integral part of Chinese culture, serves as a profound expression of beauty in the eyes of the Chinese people. However, within the Civitai platform, there has been a notable absence of a high-quality Hanfu model. In order to promote China's rich traditional culture and facilitate global understanding of its intricacies, we embarked on a journey to create an exceptional Hanfu model.

Through an extensive process of data collection and refinement, we have meticulously curated a dataset that encapsulates the essence of Hanfu fashion. Our dataset comprises an impressive collection of 32,000 meticulously sourced high-quality Hanfu images from across the internet. To ensure a top-tier collection, we painstakingly selected 20,000 images that truly embody the elegance and allure of Hanfu, discarding any that fell short of our quality standards.

The foundation of our Hanfu model lies in the Finetuning process using the SDXL architecture. Remarkably, the achieved results are of remarkable quality without necessitating additional refining steps.

Usage Guidelines: Keyword Prompt Enhancement: For optimal results, incorporating the keyword "hanfu" into your prompts can significantly enhance the model's performance and relevance.

Diverse Resolutions: Our training dataset's resolution spans from 512 to 2048 pixels due to the strategic bucket merging. This ensures excellent model performance across a wide range of resolutions.

CFG (Controlled Fine-Tuning Gain): As preferences may vary, we encourage users to experiment with different CFG values. An effective value that yielded positive results for us is 6.5, but personalized exploration is encouraged.

Optimal Steps Setting: The choice of steps is subjective and user-specific. We recommend exploring within the range of 20 to 35 steps for optimal results.

By introducing our state-of-the-art Hanfu model to the Civitai platform, we aim to bridge cultural gaps and allow users worldwide to appreciate and engage with the splendor of Chinese traditional fashion. This model stands as a testament to our commitment to preserving and celebrating the rich tapestry of global cultures through AI-powered technologies.