So I would like to ask you, if you would give me a few sec of your time:
I’m working on a project (similar to an AI influencer/Fanvue style) and I need to create a highly consistent LoRA model of one specific female face.
The biggest hurdle right now is creating the initial dataset of 25-30 photos. I only have one reference photo of the face I want to use (lora2.jpg), and I need to generate 25-30 ultra-sharp, photorealistic images where the face is exactly the same in every shot, but the pose, clothes, and scene are different.
My current video workflow (based on WAN2.2) is not suitable for generating high-quality, consistent static images for training.
What I’m looking for is a ComfyUI workflow (JSON file) that uses SDXL and the IP-Adapter to achieve this high consistency.
The problem is I have seen a few people doing iPAdapter and it was sh*t quality bad face,atd.. I tried WAN2.2 I2V where I took ref. image, create video with it and than screenshot it. but it was blur.. Now I´m wondering where should I focus?