Best base model for accurate real person face LoRA training
The user is looking for the best Stable Diffusion base model to train a LoRA (Learned Residual Attention) for a real person's face, aiming to achieve results that closely match the training images.
Why it matters
Choosing the right base model is crucial for achieving accurate and realistic face generation using LoRA, which has applications in various industries like entertainment, social media, and digital art.
Key Points
- 1The user wants to train a LoRA for a real person's face
- 2They are looking for a base model that can preserve the exact identity of the person
- 3Some models tend to average out the face instead of keeping the exact likeness
Details
The user is exploring different Stable Diffusion base models, such as Flux, SDXL, Qwen, and WAN, to determine which one handles face likeness the best for LoRA training. They have noticed that some models tend to average out the facial features, resulting in a less accurate representation of the person's identity. The goal is to find a base model that can produce results that closely match the training images, preserving the exact facial characteristics of the real person.
No comments yet
Be the first to comment