AI [AI Art] - Show Us Your AI Skill *NO TEENS*

4.80 star(s) 6 Votes

felldude

Active Member
Aug 26, 2017
572
1,724
so if you wanted to "train" a Lora of Lacey Chabert, would this page have enough pics of her to do the job
You don't have permission to view the spoiler content. Log in or register now.
?
Yes for a face trained Lora for sure, maybe enough for body shape.

Realistic vison (And maybe even 1.5 base) was likely trained with a few images of their likeness

ComfyUI_00288_.png ComfyUI_00286_.png

I didn't really like the images I was getting from the name prompt....So I trained a LORA



Example images 1Girl no name prompting U-Net training only

ComfyUI_00311_.png ComfyUI_00307_.png

Upscaled to add detial Lora value .3-5 and name prompting .35 (Lacey Chabert:.35)

Note for Auto1111 users this is not a lora prompt it is telling the checkpoint to mix in %35 percent of the name
"Lacey Chabert" training into the mix. Since the lora doesn't have the name "Lacey Chabert" it is mixing in checkpoint training with the Lora U-Net

ComfyUI_00310_.png

I'm generating face images that will pass Civitai standards and I will post the lora.

God damn it lora im trying to generate a sfw image here

+ a 4k face photo, portrait, face detail,
-text, watermark, nudity, blurry, soft

ComfyUI_00324_.png
 
Last edited:

mailpa

New Member
Sep 5, 2021
2
3
I'm really trying to squeeze something out of my CPU-only turbotoaster here, can't get a good quality image but I can play with almost all of the parameters and models to get something very custom at the end, if I'm patient enough

ashley-graham-f18_up.png
 

felldude

Active Member
Aug 26, 2017
572
1,724
I'm really trying to squeeze something out of my CPU-only turbotoaster here, can't get a good quality image but I can play with almost all of the parameters and models to get something very custom at the end, if I'm patient enough

View attachment 2941732
Based on your description of your computer I don't think you have a post 2020 CPU but if you have a "Tiger Lake" or Xeon processeser Intel has made some PyTorch extensions with impressive performance.

They also have there own integrated GPU build, I don't think any of the Web-UI interfaces use any of yet though.
 
  • Like
Reactions: mailpa

FoolsBeGone

Well-Known Member
May 21, 2023
1,043
3,137
Yes for a face trained Lora for sure, maybe enough for body shape.

Realistic vison (And maybe even 1.5 base) was likely trained with a few images of their likeness

View attachment 2941603 View attachment 2941604

I didn't really like the images I was getting from the name prompt....So I trained a LORA



Example images 1Girl no name prompting U-Net training only

View attachment 2941790 View attachment 2941791

Upscaled to add detial Lora value .3-5 and name prompting .35 (Lacey Chabert:.35)

Note for Auto1111 users this is not a lora prompt it is telling the checkpoint to mix in %35 percent of the name
"Lacey Chabert" training into the mix. Since the lora doesn't have the name "Lacey Chabert" it is mixing in checkpoint training with the Lora U-Net

View attachment 2941798

I'm generating face images that will pass Civitai standards and I will post the lora.

God damn it lora im trying to generate a sfw image here

+ a 4k face photo, portrait, face detail,
-text, watermark, nudity, blurry, soft

View attachment 2941812
She looks fantastic! She is such a gorgeous woman that I knew a Lora based on her would be hotter than lava!
 

FoolsBeGone

Well-Known Member
May 21, 2023
1,043
3,137
Using Tensor.Art I used this prompt with these checkpoints and Lora's and this was what I got. I clicked generate five times and got five different photos which was great. Any way here she is.
You don't have permission to view the spoiler content. Log in or register now.
UkDqUlKrRizdDBmDUcVxh.png
2YJ1HySDSw9884EaUITAr.png b3m521e9ZOwouZHXU6bxz.png 4ls-eavGleBoMFnicS71v.png bbw3ML4O5yM3JndNVdqKF.png
 
  • Like
Reactions: Dagg0th

Artiour

Member
Sep 24, 2017
314
1,319
View attachment 2945331
Mercedesz: Artiour, Dear..! leave those LoRas for a moment and come to bed, were are waiting.. *luv*
I have to say, you have some good level realism, I train my LoRAs to be semi-realistic semi-3d so it's possible to make realistic or 3d from them, you mastered realistic especially with this one of Kasie I still admire till today, probably got me more downloads to the LoRA
Capture.JPG
 
4.80 star(s) 6 Votes