As always in my quest to create extremely realistic models, I finally made my XL model, which was an old debt to many of you.
After someone asked me about my last Illustrious model if it wasn't a base XL model, I decided to make a comparison between those two types of models, but I think the best way to do it was to create my own XL model, which is something that many had asked me to do, but I hadn't been able to do. Little by little, during the development of it, I realized the enormous capabilities that I had left aside by concentrating on my Pony and Illustrious models, but at the same time, I think it gave me time to develop my own style that I was able to capture in this model.
What this model is good at is that it is very good as a general model for all types of generations, especially artistic and detailed ones. It is good at following the prompt and is excellent for detailed portraits.
For distant images I recommend using Adetailer if you don't use Hires.fix
These are the recommended parameters:
Since version 3 the VAE is Baked
Upscaler: Im using 4x-ClearRealityV1 upscale by 1.5 Steps 15 denois 0.35 Hires CFG 4,
Width and height: 896*1152
CFG Scale: 5
Steps: 35
The sampling method i use is DPM++2M-Karras
You may see many images that are a bit dark in the examples, but this doesn't mean that it's a trend in the model, but rather that I personally prefer dark images that don't have excellent definition.
I hope you can share as many images as possible, as this is the only way to continue improving my work.
Description
This time it has been improved with a little more realism, better colors and more cinematic images, I hope you enjoy this version as much as I do and I hope you can share your examples to continue improving, the parameters I used remain the same
The VAE is not baked, use SDXL_vae
Upscaler: Im using 4x-ClearRealityV1 upscale by 1.5 Steps 15 denois 0.35 Hires CFG 4,
Width and height: 896*1152
CFG Scale: 5
Steps: 35
The sampling method i use is DPM++2M-Karras
FAQ
Comments (3)
Tested this model V2 with and without LORA's using my favorite prompts. 🙌 It's gives amazing realistic results! ♡♡♡ Although not every seed works. It sometimes generates deformed limbs or random nudity. It takes a few attempts on random seeds to get the desired result sometimes using LORA's, but without LORA's it works very good. Guess I'm nolonger used doing retries after using Flux too much. 😜 Good job! 👍
please add as many uncommon poses as you can in next versions.
feed your model for what others can't do.
focus on human poses without destroying what you achieved (assume you will make something better in this).
Checkpoint is really good. Natural colors, natural color palette, nothing excentric or oversaturated, which is rare on SDXL ecosystem, ok skin tones, skin detail good. I can clearly see a lot of work put into it.
What it bothers me is why is the VAE not merged into it? it's for training purposes? What is the advantage of this?