This model tests a new marking method, which uses LLM to describe and mark images instead of traditional Deepbooru marking. Firstly, it can be clarified that LLM currently has hallucinations in image recognition, and there is currently no good solution.recommand weight:0.8-1prompt:waterstylellm,1girl,&lt;lora:waterstylellmV1_SDXL_lora-000006:1.0&gt;Example 1： waterstylellm, The image features a woman with long hair, wearing a shiny silver suit, and standing in a water-filled room. The woman's hair is flowing and appears to be wet, which suggests that she has just emerged from the water. The silver suit she is wearing adds a futuristic and metallic touch to her appearance, making her look like a cyborg or a robot. The water-filled room creates an interesting and visually appealing environment, with the woman standing out as the focal point of the scene.  Example 2：waterstylellm, The image features a young woman standing in front of a city skyline, with a sunset in the background. She is wearing a white jacket and has long hair. The city skyline consists of tall buildings, and the sunset creates a beautiful and colorful backdrop for the scene. The woman appears to be standing on a pier or a dock, with water visible in the background. The overall composition of the image is visually appealing and captures a moment of serenity and beauty in the cityscape.  You can see that there is a lot of nonsense mixed in the LLM, but the training effect is not very poor. You can compare it with another model of the same dataset that I trained through routine training. <a target="_blank" rel="ugc" href="https://civarchive.com/models/120521/watermiku">https://civarchive.com/models/120521/watermiku</a>

Files with SHA256: 6016dc7bbb7bb8624bf896627908c437ae7ae4d344f2202db8a32c070888593e