After testing over 2000 different PDXL models, I started testing ILXL (Illustrious-XL) models as well (including NoobAI-XL which is also based on ILXL). The general idea is the same: I only want to give a comparison of their default styles without judging them. ILXL includes many different styles that can be triggered by using artist tags so keep in mind that the default style (without artist tag) is not the only one the models are capable of. I slightly adjusted the prompts compared to PDXL but I'm using the same prompts and seed for all ILXL models.
The direct comparison of ILXL with PDXL is a bit difficult since they are different base models. PDXL has way more finetunes and developed a great aesthetic look over time. But as much as I love PDXL, I have to admit that ILXL is the superior base model. With its amazing Danbooru2023 dataset (which NoobAI even extended), it knows almost all characters with artworks on Danbooru (NSFW warning). It also has a better prompt adherence compared to PDXL and seems to be better with hands. So overall I personally prefer ILXL for Anime images.
Just like with PDXL I will only use SFW prompts for my test images. But keep in mind that ILXL is perfectly capable of producing NSFW images as well.
This comparison is not completely fair for every model, but adjusting the prompts and looking for a good seed for every model would not provide a good comparison. I download almost every new model and post the test images to their respective model pages with my test image upload account Lizardon1025.
All the prompts and settings are listed below. For some models a large amount of positive and negative quality prompts are recommended while others seems to require none at all. So I simply use two positive and negative quality prompts for all models. You can also find the ComfyUI workflow attached to this article.
Keep in mind that I'm (mostly) only testing models marked as "Base Model: Illustrious" since I set my filter this way to look for new Illustrious models. NoobAI-XL is also based on training from Illustrious as clearly stated on their HuggingFace webpage, so as long as CivitAI has no further differentiation between Illustrious and NoobAI-XL, in my opinion it should be set to base model Illustrious.
I'm listing my personal top 5 models here in alphabetical order. This choice of models is entirely subjective. I will update the list from time to time.
Latest update of favorite list:
19.02.2025
Tested ILXL models so far: ~ 1400
boleromixIllustrious_v290 (CFG ~ 3)
illusionbreed_22DSlenderV30 (CFG ~ 6)
konpaevoMixNoobaiEPS_v11 (CFG ~ 1.5)
pppanimixIL_60 (CFG ~ 4)
toonmergeIllustrious_v10 (CFG ~ 1)
The current prompts are the following:
Zelda kiss:
masterpiece, best quality, 1girl, 1boy, princess zelda, the legend of zelda: breath of the wild, standing, kiss, hug, closed eyes, outdoors
Portrait:
masterpiece, best quality, 1girl, face, portrait, looking at viewer, closed mouth
Forest Angel:
masterpiece, best quality, 1girl, solo, blonde hair, wavy hair, angel, angel wings, halo, smile, sitting, forest, white dress
Library Catgirl:
masterpiece, best quality, 1girl, solo, long hair, black hair, blunt bangs, straight hair, round eyewear, cat girl, cat ears, cat tail, standing, hug, cat, looking at viewer, library
Negative prompts:
worst quality, bad quality
All images were created with their recommended sampler, scheduler and CFG value. If there are no recommended settings to be found, I'm using these default settings:
Sampler: Euler a
Scheduler: Normal
CFG Scale: 5
Steps: 100
Width: 896
Height: 1152
Seed: fixed