Hello everyone. I am Dr. Furkan Gözükara. PhD Computer Engineer. SECourses is a dedicated YouTube channel for the following topics : Tech, AI, News, Science, Robotics, Singularity, ComfyUI, SwarmUI, ML, Artificial Intelligence, Humanoid Robots, Wan 2.2, FLUX, Krea, Qwen Image, VLMs, Stable Diffusion
Im really struggling to explain my self, for that i am sorry, but: -You describe the workflow, super nice! -But are you saying if you could just use the UI you would hit 100% correct generations, and you are "fighting the results you are getting from UI vs the json payload"?
That is what @Dr. Furkan Gözükara told me and it worked for me: Furkan Gözükara SECourses — 11/30/2023 12:16 PM ok easy to fix click start training again it will cache remaining files
use the special tokens you used in training. For example main prompt could be: ohwx man standing next to bbuk woman on a street corner. In the adetailer prompt (using face yolov8n) use: ohwx man <ohwx man lora> [SEP] bbuk woman <bbuk woman lora>. If it swaps faces then reverse this command. Make sure to set detection K of 2 so it does not swap other faces if there are more than 2. If you had a ckpt trained on both then you wouldn't have to use the <lora> portion. Either way works
ok will try. its safetensor , not ckpt . I still have 1024 pictures of both of us. is it possioble to put both of our photos and train ? If so what would be the regularization pictures ? it should be about couples right ?