General | Chinese Girl Cosplayer

---

This model does not generate NSFW content.  

This model is trained only on images of female cosplayers.  

This training and testing were conducted on Flux. Due to limited data and labeling constraints, the model cannot reproduce cosplays of specific characters. For example, it cannot replicate the cosplay of "Konan" from "Naruto". The dataset's high consistency is the cosplayers' aesthetic appeal, so the model tends more towards a "beautiful Chinese cosplayer" model rather than a strict "cosplay model".

The dataset annotations combine natural language and labels, ensuring high annotation quality. It is hoped that with low-weight usage, the model's understanding of text will not be overly distorted.

The trigger word is "Cosplay photography". Always start this phrase with the trigger word.

An example trigger phrase could be:

```

Cosplay photography, Gothic portrait of a girl with long black hair, ample bosom, lace choker, depth of field, heavy bokeh, cinematic lighting, movie colors, dramatic lighting, ultra-high detail, gaze away from the camera.

```

Since fine-tuning the T5-XXL model wouldn't be meaningful, this training did not involve fine-tuning of T5-XXL. Therefore, you can directly write trigger phrases purely in natural language. It is recommended to use some visual-language models (such as GPT-4o) to infer images and generate trigger phrase suggestions.