ZIT appears to deliberately censor male genitalia, sometimes even compromising anatomical accuracy.

Previous training attempts consistently failed, leading me to suspect deletion biases were already present in the LLM.

Therefore, I deliberately selected a keyword non-existent within the model for testing.

The selected images comprised roughly equal proportions of anime and photographs.

None were real photographs; instead, I generated approximately 80 images using my own Unstable Ink Dream Illustratious model.

Version V1 used only keywords without subtitles.

ZIT中似乎故意刪減了男性的下體,有時候甚至影響到正常的解剖學

之前訓練一直不成功有點懷疑從LLM那邊就已經有刪除影響了

所以刻意選擇了一個模型內不存在的關鍵詞來做測試

因為選用的圖片動漫和照片類大約1:1

都不是真人圖片而是用過去自己的unstableinkdream illustratious模型生成約80張照片

V1版是不加字幕僅關鍵字的