The development of this model is based on my personal interest and amateur practice, and it does not have professional training or deep knowledge. Therefore, there is still significant room for improvement in the output. This model was trained using about 2000 SFW and NSFW images related to homosexuality that align with my preference for male appearances. Theoretically, this model is better at generating images of men with elements that exist in reality, but it does not mean it cannot generate images of women or other styles, or images containing fantastical elements, although the results of these images might not be as expected.Testing was conducted at a resolution of 768*1024, primarily using the DPM++ 2M Karras sampling method. About 20 steps can produce decent results. I strongly recommend using ADetailer for facial refinement, while I advise against using Hires. fix for image enlargement. When the generated images of men are not satisfactory, trying keywords like "1man", "handsome face", "sexy guy", "hotbromance", etc., might help. Especially when attempting to generate images of men with non-realistic features (e.g., red eyes), using many and complex prompts might result in unexpected outcomes (such as blue eyes). In such cases, using the unexpected results as a reverse prompt (adding blue eyes to the reverse prompt) might improve the issue.Since the material used for model training mostly consists of men with a more muscular physique, unless the generated characters consistently have a slimmer body type, it might not be necessary to add terms like "muscular" to avoid generating results with overly inflated muscles or severe anatomical distortions when generating characters stereotypically perceived as having a muscular physique.這個模型的開發基於我的個人興趣和業餘實踐，並不具備專業的訓練或深厚的知識。因此產出的效果仍存在很大的改善空間。此模型使用了符合我對於男性外觀偏好約2000張包含同性戀相關的SFW和NSFW圖片來訓練。理論上此模型對於生成帶有現實存在元素風格的男性影像效果較佳，但並不意味著它不能生成女性或其他風格，或包含非現實元素的圖片，儘管這些圖片的效果可能不如預期。以768*1024的分辨率進行測試，主要採用DPM++ 2M Karras作為取樣方法。大約20步就能產生不錯的效果。我強烈推薦使用ADetailer進行面部修復，同時不建議使用Hires. fix來放大圖片。 當生成的男性圖像效果不盡如人意時，可嘗試使用如 "1man"、"handsome face"、"sexy guy"、"hotbromance"等，可能會有所幫助。特別是於嘗試生成具有非現實特徵（例如紅色眼睛）的男性圖像時，尤其當提示詞較多且複雜時，可能會出現非預期的結果（如藍色眼睛）。在這種情況下，將非預期結果作為反向提示反饋給模型（在反向題詞加入藍色眼睛），可能可以改善問題。由於用於模型訓練的素材多為身材較健壯之男性，除非產生角色的身材結果均為較苗條之體態，否則對於生成刻板印象上已具有較健壯的體態角色類型時，可能不用加上"muscular"一類的題詞，以避免產生肌肉過度膨脹、解剖學嚴重變形的結果。

Files with SHA256: 2dcd9887ccaf5c2c3ef4980fe5eff6cee0748238ef2e66a4dbe33a0216002f43