Internet celebrity face model
V1







1. This is a photo of liangliang woman. The weight is 1. If the trigger word is included in the description, it is recommended to start directly with "This is a photo of liangliang woman." Generally, trigger words are not necessary for face models. Using them can help stabilize the face better.
2. Using both "lora" created by me together with a weight of 1 may result in contamination similar to overfitting. There should be no issue when using "lora" created by others together with mine. This is caused by the trainer, and I have studied it for a long time but have not been able to identify the problem.
3. Download the image and drag it to comfyui to get the workflow. I have been iterating on this workflow all the time. Besides using the set node to avoid messy connections, I try to ensure that the workflow's routing logic is understandable and easy to modify later. It should be quite useful if you can run it. For the convenience of laziness, I used the meow-hand reverse prompting words, which have a similar effect to Joycaption but much faster. I used the grok API to polish the prompting words (the polished prompting words can be written in 2.1 or can be reverse-promoted). No magic or money is required. You can directly tell it your requirements in Chinese, which is stronger than any local model. By enlarging the sparse chunks (4.3), the original composition can be maintained, adding details and clarity by increasing the steps, which can also be used to remove stripes caused by the flux large image (magic is required for node installation).
Disclaimer: This model is for learning and communication purposes only. Please comply with the laws and regulations. It is forbidden to merge, reprint, or generate NSFW images. The consequences of using this model are the responsibility of the user.
Disclaimer: This model is for learning and communication purposes only. It is forbidden to merge, use for illegal purposes, generate NSFW images. The consequences of using this model are the responsibility of the user.
