<a target="_blank" rel="ugc" href="https://gayeroticvideoindex.com/performer/18916">Peter Berk</a>, of the 90s era wanker fame! Don't know who Peter Berk is? Uh helllooo.. he's the reason why almost every homo wanted to get their hands on a tape of <a target="_blank" rel="ugc" href="https://gayeroticvideoindex.com/video/521">Young, Hard &amp; Solo #3</a> back in the mid-90s! I mean, besides seeing Simon Rex yank it. lolThis was more of a challenge on building a decent LoRA when there's very limited amount of quality source material and that retains a high degree of facial accuracy fidelity but also fairly flexible. I know there are approaches to build LoRAs with a limited amount in the dataset but they tend to be very inflexible and/or don't retain a high degree of facial accuracy outside of a few poses.The other thing I wanted to do was test out some automated approaches to get to a decent dataset making it a low effort/high yield ratio kinda thing. That last part still needs some work but I'm getting there.Anywayz - main trigger word: peterber

Files with SHA256: a3929f9c055e10d7e9aa3706c38213f7c923f0c4b7785e9003292a4367b40835