That is quite the difference, could it be all the ADetailer used in the orignal prompt
Synalon
Member
- Jan 31, 2022
- 228
- 668
I use Adetailer, but also refined it in img to img based on the original I created in text to image, and then I keep the original in controlnet as a reference so the pose doesn't change to much.
I also get totally different poses based on the model, so a different model may make a different image as well.
I'm using Juggernaut btw.
I'm also using clip skip 2.
This is what I got with a different model. Its not bad but its not what I want.
If you want to keep the pose load the image into controlnet - openpose, load the preview and tell it to use preview as input.
I also get earlier "concepts" bleeding into a new concept so I tend to completely reload SD for each concept.
Last edited:
Sharinel
Active Member
- Dec 23, 2018
- 613
- 2,586
Now I had to try out juggernaut.
One is refined with Juggernaut the other Realistic vision merge
One is refined with Juggernaut the other Realistic vision merge
Sharinel
Active Member
- Dec 23, 2018
- 613
- 2,586
Adetailer only really changes the face, it doesn't affect anything else - deffo doesn't change you from a dungeon to a cliffside and pull your skirt up
Synalon
Member
- Jan 31, 2022
- 228
- 668
Juggernaut has a negative embedding, so I always add that, and its got a VAE it likes as well.
Synalon
Member
- Jan 31, 2022
- 228
- 668
Juggernaut is the name of the model I use within Stable Diffusion to generate the image.
PandaRepublic
Member
- May 18, 2018
- 215
- 2,167
Study Sessions With Luisa
You don't have permission to view the spoiler content.
Log in or register now.
Digitalface
Member
- Dec 20, 2021
- 120
- 2,524
Start your day right with daily stretches.
Last edited:
Why is every.....SHUT UP MEG
You don't have permission to view the spoiler content.
Log in or register now.
Last edited:
FoolsBeGone
Well-Known Member
- May 21, 2023
- 1,043
- 3,137
FoolsBeGone
Well-Known Member
- May 21, 2023
- 1,043
- 3,137
Also I have to ask this is directed at Loserriot. Dude what are you doing in this thread if you are going to facepalm every Picture why don't you go find something you like instead. You have facepalmed some truely epic portraits and I can't figure out why you are hating on every thing. No hate here I just can't figure out why someone comes into a thread just cause drama.
I'd ignore but "artist" feed off attention and reactions so he is still feeding the monster....Its also possible that is just their thing or emote illiterate.
FoolsBeGone
Well-Known Member
- May 21, 2023
- 1,043
- 3,137
All of the renders look good but THIS one is *chefs kiss* You perfectly captured the essence of Glory. She is very unemotional (on the surface) due to trama in her past. But there is a fire inside of her and I think that this picture gets it, and encapsulates it perfectly.
Thank you, other then look at what art existed already for the character. (Not much that I found) I'm not familiar with the lore, or game. (Their where some WIP screenshots of a 3d model that would have helped but I couldn't find it)
I had to hand paint the color in around the eyes on that image, it was one of the ones I used to train the second iteration of the lora
Today I have been working on redheads:
Last edited:
FoolsBeGone
Well-Known Member
- May 21, 2023
- 1,043
- 3,137
I don't know if it would help but I can screenshot some of her from the game then post to see if it helps. I can honestly say though I am super impressed by what can be done with the AI. Scary perhaps but so is anything in the wrong hands. In the right hands it can bring about some amazing things.
You must be registered to see the links
The type of game it is based of the screenshots I have looked at wouldn't train a character lora, I was able to train the AI with AI.
Its pretty common to do, you just use the best images to refine, or in this case I used them to train another lora as the first was ok but needed help.
Minimal prompting is needed to get the image below, for a character lora that is ideal as you can always dial it back to allow for different details.
Sometimes the seed will change things up too, like in this case I asked for a tube top and got:
And then if it can't produce different poses with the details it is pretty useless also (If I where to fully refine the lora I would need to run a batch of a few hundred different poses and use the best ones)
Last edited:
I couldn't get XL to put the startrek emblem in so I gave up
