Elba Notexist

Who is Elba? Do you recognise her? You don't. She's a creation of Stable Diffusion and is here for your use in your prompts.

For best effect, use Adetailer.

In my continuing battle to get less prettified characters, I decided to try to use the IP-Adapter Controlnet plug in to add a face to a dataset. This worked in a much more accurate way. This actually trained with exactly the same face (though a different body) than Donna, which says the impact the faceswap method has. It uses IP-Adapter and Controlnet to make sure the placement is correct. The problem with this method is that while Reactor will map the expression of the dataset to your face, this will just add the standard face. This means that I've used Faceapp to add some smiles of various sizes to the figure, just so it doesn't become one of the models which are deeply inflexible. You also, when doing a batch, will likely swap between 512x512 and 1024x1024 to get the inpainting you want, and then re-size. There's images with and without glasses in the dataset, so it responds well to that prompt.

In other words, it's a bit more effort (the Reactor method is incredibly easy) but the only way I've found to create embeddings like these.