V0.6 does what I intended the model to do as this is my main model. You may have to put in your negative prompt "Sepia" if you don't want it, as it is what is causing the less colored sort of look on V0.6.
Feel free to publish images made from this. I like to see posts that aren't only mine and I will usually like the posts that aren't mine if I like the images or if the images does what the model is supposed to do.
Feel free to tip, like or follow the model to show support to the model and me. Model is available on tensor.art for site generation. (https://tensor.art/models/822018146149058892). V1.0 and V0.8 file is embedded with the workflow made to make it. Old V1.0 file is now replaced as of 3/24/2025.
Recommended V1.0 Settings:
Steps: 30 / (Probably lower, like 25 steps, if you want to upscale or whatever)
CFG: 7+
Sampler/Scheduler: Euler A Normal / DPM ++ 2M Karras (Haven't tested outside of this)
Sometimes a Face Detailer, Upscale, or rarely Hi-Res Fix.
Recommended Negative: worst quality, old, early, low quality, lowres, signature, artist name, signature, username, logo, bad hands, mutated hands, bokeh, simple background, jpeg artifacts, simple background, very displeasing, displeasing, bad anatomy, artistic error, signature, artist sign
Recommended V0.8 Settings:
Steps: 30 / (Probably lower like 20 to 25 steps, if you want to upscale)
CFG: 7+
Sampler/Scheduler: Euler Normal / Euler A Normal / DPM++ 2M Karras / (Haven't tested outside of this)
Sometimes a Face Detailer, Upscale, or Hi-Res Fix.
Recommended Negative: worst quality, old, early, low quality, lowres, signature, artist name, signature, username, logo, bad hands, mutated hands, bokeh, simple background, jpeg artifacts, simple background, very displeasing, displeasing, bad anatomy, artistic error, signature, artist sign
Recommended V0.6 Settings:
Steps: 25
CFG: 4 - 6
Sampler: Euler / DPM++ (Haven't tested outside of this)
Scheduler: Normal / Karras (Haven't tested outside of this)
Sometimes a Face Detailer or Hi-Res Fix.
Below is the Workflow (Save File is Embedded) used to make V0.6:
V0.1 for prompt coherence isn't great, it's bad overall I'd say.
Below is the Workflow (Save File is Embedded) used to make V0.1:
Thumbnail was made using a 1060 Series GPU, so it could be better, if you have a better GPU than me.
⚠️ IF YOU'RE GOING TO STEAL MY (https://civarchive.com/user/Arctenox) MODELS OR POST THEM SOMEWHERE ELSE, MAKE IT SO THE DOWNLOAD IS ENABLED, IF IT'S NOT A BOT UPLOADED IT.⚠️
https://civarchive.com/models/1149979
Description
I had to restart the merging process with this version. This is mainly NoobAI and Illustrious Merge now with a few loras of mine and many others that aren't mine at a attempt to create something unique. This version is much better at prompt adherence too.
This version has a better style and better prompt adherence. This version is also not great at porn.
This also only has 1% Model/Clip of the V0.1 file, so this version is pretty much mostly a New NoobAI/Illustrious Merge with some slight Pony.
Additionally the VAE is baked in, it's the "SDXL VAE" for those who are wondering.
The block merge in this version was with Badromance - Pony V0.9 (https://civitai.com/models/1141423/badromance-pony)
FAQ
Comments (7)
how good it is able to support Pony and xl Loras? can it support this Pony Lora? https://civitai.com/models/365388/saiyans-pony-xl-son-gokuson-gohanson-goten-vegetatrunksbrolybardock-napparaditz-ozaruvegetogogetagotenksgotrunkssayansssj2-ssj3ultrainstintoultraego
V0.7 Model works with some Pony LoRAs, and XL. I have no idea about character LoRAs, also complicated things don't work yet with this model, so if it doesn't follow the prompt it's cause the model doesn't understand, yet. I tried fixing that as much as I could before I uploaded it.
Edit: I'm better at making Anime, Abstract and Quality LoRAs it seems as well...so uhh... sorry if the checkpoint model is bad I tried making it do semirealism by just merging, I'll be merging in LoRAs for V1 to get it some coherence and what not but also try to make it 10%-ish more noobai/illustrious. Although model probably isn't as bad as I think due to me only being able to test it on Tensor.art and Low VRAM mode on ComfyUI as I have a 1060 3GB GPU, not the best thing lol.
The next version of the model which is almost done is way better in prompt adherence although VERY LITTLE of Pony is in it. So I have no idea if Pony LoRAs will work with it.
But XL should still work with it, it's noobai/illu mainly. Most XL LoRAs work on Both Pony, Illu, and NoobAI as those three are finetunes of XL. Although I don't know if it'll look any good with the next version (V0.6) lol.
@Arctenox wow great, the model is not bad at all the thing you are doing with 1060 3GB is insane and I know Pony is based on SDXL but hard to do things with it when it comes to making it work with others btw thanks for the effort you models are amazing and yeah if you are good with anime then may you please train a Lora for dragon ball super anime characters? Illustrious would be amazing
@krigeta I suck at character LoRAs, so I've also discovered due to making My Own OCs', LoRAs. Otherwise I'd do em more often oof.
I seriously need to learn how to do Character LoRAs properly.
@Arctenox sure dataset is the only thing left otherwise you are a pro
@krigeta On a different note if you wanna try V0.6 of the model go ahead, cause I'm obsessed with it atm lol.
If you haven't already.