• To improve security, we will soon start forcing password resets for any account that uses a weak password on the next login. If you have a weak password or a defunct email, please update it now to prevent future disruption.

3D-Daz Fan Art Holiday Island: Art Assets and Fan Art Thread

MarsUltor

Active Member
Aug 2, 2018
764
1,687
AI doesn't always do what I want. What I said was "huge". :cool:
Indeed. Building effective AI prompts is as much art as it is science. Sometimes you have to "double (triple, etc.) up" on prompt descriptions with words like "bombshell", "busty", "buxom", etc.
 
  • Like
Reactions: -=Cazz=-

Jimwalrus

Active Member
Sep 15, 2021
820
3,093
AI-generated Ivy on a beach:
00012-1953402741.png
00015-3739927186.png 00020-2496541588.png

LoRA is (not mine). The same user also created them for Amy, Eva, Jennifer, Jessica, Heather and Yvette. Some work better than others - Heather seems good, Yvette and Jennifer not so much.

Edited to add:

These are the best of a test run of . As I said, I feel it's not as good a representation as Ivy's.
00130-1227528096.png
00124-1227528090.png
 
Last edited:

Jimwalrus

Active Member
Sep 15, 2021
820
3,093
DAMN!!!! Those are almost DEAD-on!!!
Thanks!
Allowing for the more photo-realistic faces that Stable Diffusion can make, for Heather, Eva and Ivy the 'hit rate' is about one in three*, another one in three are good but have wonky hands / feet and the final one in three are subtly "just-not-her". The latter case is hard to define, but obvious to those of us familiar with the girls (every square inch of the girls...:ROFLMAO:)

seems to have a much lower hit rate - not had a good one yet.
I'm working on her and at the moment (oo-er!)
You don't have permission to view the spoiler content. Log in or register now.

*and about 1-in-20 for the "DAMN!!!" :cool:
At ~2m30s per image, to get that best 5% it's probably slower than generating them in DAZ then in-painting their faces with Stable Diffusion. But I've no idea how to use DAZ, so there we go.
 

MarsUltor

Active Member
Aug 2, 2018
764
1,687
Thanks!
Allowing for the more photo-realistic faces that Stable Diffusion can make, for Heather, Eva and Ivy the 'hit rate' is about one in three*, another one in three are good but have wonky hands / feet and the final one in three are subtly "just-not-her". The latter case is hard to define, but obvious to those of us familiar with the girls (every square inch of the girls...:ROFLMAO:)

seems to have a much lower hit rate - not had a good one yet.
I'm working on her and at the moment (oo-er!)
You don't have permission to view the spoiler content. Log in or register now.

*and about 1-in-20 for the "DAMN!!!" :cool:
At ~2m30s per image, to get that best 5% it's probably slower than generating them in DAZ then in-painting their faces with Stable Diffusion. But I've no idea how to use DAZ, so there we go.
For the wonky hands, can you use a negative prompt to avoid those ("malformed hands:-1.0" or something like that)?
 
  • Like
Reactions: Jimwalrus

Jimwalrus

Active Member
Sep 15, 2021
820
3,093
For the wonky hands, can you use a negative prompt to avoid those ("malformed hands:-1.0" or something like that)?
Believe me, I've got all the negative prompts and negative embeddings designed to minimise wonky hands already in there.
The reason Stable Diffusion struggles with hands (and I'll keep it brief because this isn't the SD thread) is simple - it was trained on 1:1 ratio images (billions of the fuckers). Human beings are not 1:1, so extremities tend to get cropped out. It's equally bad at feet and lower limbs for the same reason.
The SD community is trying lots of potential fixes for this, including creating SDXL which was trained on 2:3 ratio images.