Wetlook World ForumCurrent time: Mon 13/01/25 07:24:20 GMT |
Message # 100861.2.1 Subject: Re:[AI] Quick update and progress notes! For Quiet271 and Oldbewet Date: Sun 05/01/25 14:34:49 GMT Name: Hudson |
Report Abuse or Problem to Nigel at Minxmovies
|
Hi and thanks! I am familiar with that LoRA, but I never got results I was happy with. maybe it's just a matter of taste. what are some of your best works with that LoRA? I might need to give it another chance. |
In reply to Message (100861.2) Re:[AI] Quick update and progress notes! For Quiet271 and Oldbewet
By James Adrian - Sun 05/01/25 02:56:01 GMT Dont understand the problem. Me and my friends have already created 1000 wetlook photos with flux. Its not that much complicated. Just need the right lora
https://tensor.art/models/769900487538618160
|
In reply to Message (100861) [AI] Quick update and progress notes! For Quiet271 and Oldbewet
By Hudson - Sat 04/01/25 21:29:22 GMT https://imgur.com/a/NU9sXKS
First off, thanks to Jake, Quiet271, and Oldbewet who responded to my last post!
Quiet271: Thanks for your questions about my process...I hope you'll understand that I want to keep some of that under wraps for now, mostly because I'm not so confident in the output thus far....I'm really only sharing it because I think people like you interested!
- Lesson one I learned: the focus has to be Flux now. And we still have __many__ limitations with Flux. It still doesn't understand what wet fabric is. Or what it means to sit in a pool with your clothes on. Stable Diffusion will always have a place in my heart though LOL.. - Lesson two: you have to train the models yourself if you want the best results. The other models did a great job, but the results from my own model I think are a little better. But the huge thing is that I get passable results WAY more often. Like 1 out of 2 generations is OK, whereas it used to be like 1 out of 40 with the third-party LoRAs.
Oldbewet: you have an eagle-eye for details. Let me know what you think, if you can, and I'll try to improve the model. I too noticed the bee-stung fingers of one of the models in the last update I made! hahah. But in a way, sometimes anomalies like that are more real. Less produced and "perfect".
My (the?) problem is that AI still cannot synthesize. The best results come from scenarios on which the model is trained. I think the last photo, of the girl in the bathhouse (?) with the black-and-white shirt probably turned out the best. No surprise why, I trained my model on professional content. Which kinda sucks because the situations are super boring, but usually the photography is the better than candid photos. Ah!! what if I trained on art sessions that weren't from wetlook producers...there's a thought for next time! I tried to make a pic of a girl in a phone booth, only it's raining inside the phone booth. The models didn't really "get" that.
Thank you all very much!!!!!!!
|
Report Abuse or Problem to Nigel at Minxmovies
If you enjoy this forum, then please make a small donation to help with running costs:
(you can change amount)
|
[ This page took 0.015 seconds to generate ]