I was inspired by some few images I saw here on CivitAI, from back in 2024, and for SD 1.5 no less, so I'd like to give tribute to that user and, to that end, I put the model in resources, it's called "concept bdsm frame".
This will structure a machine in a particular fashion to secure a person.
Krea 2 can make all sorts of interesting machines, though they can differ quite a bit from each run, but with a little help, and not much training, we get enough consistency to call it finished.
With something like this, when prompting the item, you could waste more than half your tokens just on the machine alone, which is doable but such a waste. This is why some LoRA are desirable as apposed to pure prompting when the model can effectively produce the same material, even better sometimes.
I love machines like this and if you do too give me a heads-up and if you'd like to see something specific I have nothing but time on my hands. Well, limited, but still pretty much just time O.o . If you see me missing for more than a few months then it probably means I ran out of that stuff but I'm glad this AI stuff came about while I still had some left.
Description
FAQ
Comments (6)
Cool! You could maybe do something inspired by these if you wanted to; https://civitai.red/models/1256320/permanent-bondage-frame-il?modelVersionId=1416430 & https://civitai.red/models/611067/permanent-metal-framebent-over?modelVersionId=1804028
Nice! Interesting challenge and exactly what I'm looking for. That goes on the list. I've got two LoRA to finish before that and then I'll start exploring dataset material.
Thank you! \o/
It's a great concept, but Lora's version is a bit muddled—it's overfitted.
Over-fitting of a LoRA isn't a singular "defect". I'm assuming you chose to say something because you'd like changes, but I wouldn't be able to know which changes to make without clarity.
I happen to know that it's skillfully over-fitted in a specific area, though I don't think I hit it hard enough because it's not giving me the specifics I want, and, instead, it's generalized, which is actually the opposite of over-fitting.
If you'd like to point out anything particular I'd be happy to have that exchange.
I smell machine oil.
@Tarterbox To be honest, LoRa outputs look a bit too much like 3D renders or CGI, which is pretty sad for something as powerful as Krea 2. Since there's so little real-world data available for these styles, it’s inevitable that the model struggles to learn them correctly.
@alex9441
It's what I intended. I fun the way I want, you fun the way you want. I don't need a rule to enjoy my hobby.