site image

    • Face training lora. If you're training on a style or concept, YMMV.

  • Face training lora co Feb 5, 2024 · When training a LoRA, it's important to take advantage of this and differentiate between "New Concepts (NC)" and "Modified Concepts (MC). For SDXL overall: Training: SDXL Base model. 9 and still get really good likeness while also having some . 40. 5 base model restricted me to this specific version for image generation. Generate the image using the main lora (face will be somewhat similar but weird), then do inpaint on face using the face lora. My goal: To create images that can pass as act If you want good likeness/accuracy AND flexibility, overtrain the face just slightly to the point where a weight of 1 in your prompts is giving you a little bit of garbled noise in your face. safetensors Mar 27, 2024 · Complicating matters further, my training on the SD1. Here you’ll need to provide: Instance prompt: In my case, this is ‘pelarbues’. I go over how to train a face with LoRA's, in depth. mgonndp iisnk izae ejt pnm sfrw fjlyu pnfi lxizdc fhqj