Updated: Nov 17, 2025
character
Zoe is the result of this workflow, along with these embeddings.
In my opinion, it's much more difficult to achieve good images with the Lora than with the embeddings, but that's certainly a matter of personal preference. The advantage of the Lora is undoubtedly that it can be used with the Civitais On Side Generator. This doesn't work with the embeddings.
As far as I've determined in my tests, the Lora requires very detailed prompts regarding image quality. The Lora strength should be between 0.5 for close-ups and 0.7 for wider shots.
Image generation without post-processing, apart from high-res fix with remacri original, 2-4 steps, 0.1 denoising-strengh, upscale by 1.4
Feel free to create your own lora with my dataset. If i like it, i will link it.