Face Consistency without training/LORA Seeking face consistency? It's possible in some tools, but the mothership called Stable Difussion has an excellent plugin that does an impressive job. Let's start with Miranda

Miranda was made in Midjourney v6 with a fairly simple prompt, just to get a good head-on face.

In Stable Difussion, install Reactor as extension (link in last post). With Reactor, you can use a single image as source, a folder with multiple images or you can build your own tiny model from one source image which you can save. I used this single Miranda-image.

After SD generates the image, the face will be swapped by Reactor. You can upscale the image within Reactor too and it's also compatible with multiple faces in one image. Anyway, Miranda went to office, the gym, got married and had her honeymoon.

Disclaimer: the results are impressive already. It failed a bit to copy her tiny freckles (yes, v6 freckles) but overall I'm impressed. Link from Github: https://github.com/Gourieff/sd-webui-reactor…