tobbelobb

Building an AI waifu empire

269 Watchers320 Deviations

6.4K

Pageviews

Watchers 269

See All

avielthe

AAAGameArtStudio

figsvams

0w3nx7day

baguskkmk

sexygrl1991

ParhamKhalili

nilijfjdhj

robby308

Sibnylu

rldal

Tobiy

Watching 26

See All

KokoroAiArt

Prelfgi

efastcruex

BoltedOnGirlsAI

nikkianders11

Elnoagallery

AiSexyGirls

Amaraeva

BlackPink-YG

ClintCearley

Nanaki31

SecondHaven

seekwithin

Wallpaper-Zone

RemiCorner

About tobbelobb

Japan
Deviant for 16 years
He / Him

Badges

My Bio

Next to a full time job and life in general there isn't much time left for being creative, but AI has made the process short enough that I have time to make something presentable on occasion. I work on whatever I feel like at the time, so my content is probably a mixed bag. You might find NSFW, but there probably won't be any sexual content. Another thing you won't find here is furry stuff. The furthest we go is a cute pair of cat ears.

Back in the days when you could pick your deviant title, I used to be "deviant of many talents". There might be some very off-topic posts now and then, but that's nothing new, I've always been all over the place.

Needless to say, all humanoid models are created to be at least 18 years old. If you spot something that seems over the edge, let me know and I will probably remove it.

Favourite Gaming Platform

Other Interests

Motorsports

Donation Pool

Core Kickstarter

It's over 9000.

9001

Goal

Donors

Status

0/9001 points

Donate Points

Posts

See All

tobbelobb

Mar 28, 2024

Opened a simple Commission, check it out in the Shop section.

Join the community to add your comment. Already a deviant? Log In

tobbelobb

Nov 8, 2023

I decided it's for the best to avoid doing NSFW of real people, and removed one upload accordingly. Some nudity should be fine, but I'll reserve that for fictional characters.

Join the community to add your comment. Already a deviant? Log In

tobbelobb

Sep 10, 2023

Lazy Training for Personal Likeness, PoC

0 min read

LoRAs are all the hype. They are lightweight, they are quick to train, they can do just about anything, and they are supposed to get along well with SDXL. If you can articulate the problem, there is probably a LoRA for it. However, this post is not about LoRAs, this post is about training Hypernetworks on top of Embeddings. Now, why on earth would you want to do that? For me, it all started with an experiment with Embeddings. I found a decent guide for training embeddings in Automatic1111, and gave it a shot. The results were inconsistent. The same Embedding on the same prompt gave wildly inconsistent likeness depending on the seed, ranging from very good, to nowhere at all. I wanted to quickly explore my options within Automatic1111 before I moved on to Kohya to try LoRA training, but information on Hypernetwork training is very scarce, and my results were poor. I was just about to give up when I stumbled upon a Reddit post that talked about "using celebrity instance tokens for