This series of the models was started (v0.4) from DucHaiten-Pony-Real (beta) by DucHaiten.
Version v1.0 was trained using a LoRa mode in One trainer. Next version are trained in Fine Tuning mode (full SDXL training).
Up to v1.0: Photorealism, lighting and anatomy related LoRas from CinEro series are integrated into this model in order to amplify cinematic render.
Goal of training was to improve to stability of anatomy and make the picture more realistic and moody / cinematic.
Version 1.1: shifted to fantasy / sci-fi themes with less realistic and more contrast / unstable rendering.
Version 1.2 [SOON]: is v1.1 further development with shift to horror / tentacles theme.
PS: be responsible and don't abuse this unrestricted model.
Description
FAQ
Comments (21)
Great model - works very well. Thanks!
"We stay on the shoulders of giants..."
An excellent improvement over Duchaiten!
Sort of takes Duchaiten and smooths over its lighting/surrealism, improves anatomical consistency, and tones down the (previously overzealous) DoF effect.
Great work! <3
I just decided to make a Meostigmat LORA with lovely bokeh for my big pleasure and avoid putting the bokeh for everyone else in base model... Not all persons are crazy about creamy bokeh.
OMG ...
with 4 Loras and hiRES-fix you can make any model look good :D
1. Any examples with SD1.5?
You can find the comparison grids with the same techniques applied to different models with different results.
2. The purpose of image posting is to show vanila and high end possibilities for different audiences.
3. Can you suggest me a way to post the high quality entertaining images without appearing them in the thread of this model, but keeping the meta for the community?
BTW, latest images created without LoRas in reso 1400*1080.
No HiRes Fix, no Ultimate Upscaler or smth. Which images you mean?
its up on the model .. yes OC high quality maybe catch more people, but if they try it ...
and your prompts ok, i get one, dont follow very good ... see in galerie my fix
@sevenof9247 PONY's CLIP is very contaminated. It is not a surprise that prompt adherence isn't ideal. Feel free to compare my model with other models you thing better. I will glad to see its weak points explicitly to improve it in future.
As for adherence, I am experimenting with the Omost Pipeline which improves the adherence greatly.
Thinking about writing a post with guide.
https://civitai.com/posts/4032533 No LORAs
https://civitai.com/posts/4032569 No LORAs and LowRes
@homoludens ... i dont use compfUI ^^
you aware that all new stuff has not pos-prompt?
This is your prompt without shitty sampler and at low res, low steps and no HiResFix
https://civitai.com/posts/4032941
And this is an example with my cinematic additions. They shift the image into realism. So no fancy ears.
@sevenof9247 You can just press the button with nodes in meta information and paste the json into any editor to se there is no LORA nodes in workflow used... I'm not hiding anything. And you should know that regular prompts from A1111 can be easily edited in CivitAi loader to remove the LORAs.
@homoludens but this prompt include "alien", "long ear", "evil eys" so i want see that if i call it :D
@sevenof9247 1. Do you have a comprehensive definition of "evil eyes"? Is it just a trashy eyeball without properly drawn cornea? or what? :D
2. Look at her eyes closer. Eyes narrowed like a predator's. Is it an eyes of mother loving her child or it is more like an eyes of sophisticated woman having a knife for your stomach hidden somewhere?
@sevenof9247 BTW, is there a LoRa specializing on Evil Eyes? Cat eyes, dragon eyes, snake eyes?
Amazing model, testing without any lora and just some embbeding and can get great results
Any plans to update this now that Duchaiten v1.0 came out??
I'm thinking about this. First, I need to make in depth comparison. Second, I need to decide what points in DucHaiten's model I can improve. And which way I can shift it to get bigger satisfaction myself.
This will take some time. I am temporarily out of business...
Washed out at CFG 3, but I like what is generated. Is there a workaround?
CFG 3 and lower gives very low contrast on the most models. You can try using the higher CFG (I tested this version with 4.5..8.5) or use LCM sampler (CFG 3 is closer to regular for LCM)