- Forum Clout
- 794
As a continuation of this thread, I created a simple LoRA model of Nana, which can be used for generating better pictures of her. It essentially adds a bunch of her images on top of a much larger image model, so instead of prompting with "Anthony Cumia" and getting something that looks like Sammy Davis Jr. or a kabab vendor, you'll get a much more ethnic and accurate image. It's a super basic model that I put together quickly, so the results can be way off. I've only trained it with the bloated pics from 2016+. I'll need to refine it with clearer pictures with more accurate training descriptions, but so far it's promising:
These are the raw output images from SD using the RealisticVision2.0 model with the UniPC sampler, Sample steps = 50, CFG = 5.5, LoRA weight = 0.8, Restore Faces OFF.
If you dont already use automatic1111's Stable Diffusion UI and have a decent GPU, go to this rentry page, follow the installation steps, then follow the steps under "LoRAs (Low Rank Adaptation)"
Download the LoRA here
Also, if I'm going to improve it, I'll (unfortunately) need more pictures of her cobblestone face. Most of the pics I used were mid/low resolution, so if anyone has links to better quality images, PM me and I'll add them and re-train.
These are the raw output images from SD using the RealisticVision2.0 model with the UniPC sampler, Sample steps = 50, CFG = 5.5, LoRA weight = 0.8, Restore Faces OFF.
If you dont already use automatic1111's Stable Diffusion UI and have a decent GPU, go to this rentry page, follow the installation steps, then follow the steps under "LoRAs (Low Rank Adaptation)"
Download the LoRA here
Also, if I'm going to improve it, I'll (unfortunately) need more pictures of her cobblestone face. Most of the pics I used were mid/low resolution, so if anyone has links to better quality images, PM me and I'll add them and re-train.
