Hundreds of Lulus: Evaluating and Scoring Illustrious Checkpoints 1-20

Intro

Amongst other things I dislike about Civitai at times is that with a flood of posts and comments you can easily forget/lose something if you don't get to it right away. Last week someone posed a question, or made a comment, and I promised them I'd write an article in response. Of course, now I can't remember what the exact question was, who asked it, and where to even start looking to track it all back down.

In the end, the question related to Illustrious Checkpoints and the person noted they saw a trend in which checkpoints were used more frequently than others. So I figured I'd go WAY overboard and try to do a review of the Illustrious Checkpoints as of today the 16th of September, 2025.

I know that there is a wide range of skill levels on Civitai, so I'm going to try and set this up more for the person that might just be exploring the site for the first time. I've only been on Civitai for about 9 months and in that time the stats tell me I've generated just over 31K images on site. By no means is that stat alone super impressive when some creators are over 1M, but I'm pretty confidant that a good portion of those images have been of my go to all around character Lulu. While I've created all manner of images with her, the overwhelming majority of those images were created using Illustrious. There was a point where I was Pony-curious, but that was more when I was in college or I occasionally drink to much.

Base Prompt and Settings

So to provide you dear reader with a breakdown of the Illustrious Checkpoints I'm going to step through them using the exact same prompt and settings

Additional Resources:
Lulu_Prime

Prompt:
character sheet ((front view, 3/4 side view)), highly detailed, 
best quality, masterpiece, 1girl(slender, femboy, lulu_prime, 
hazel eyes, pink hair, dual ponytails, small breasts, 
wearing white button down shirt with collar, men's black neck tie, 
red plaid microskirt, knee socks, black patent leather Mary janes),
intricate details

Negative Prompt:
lowres, error, cropped, worst quality, low quality, jpeg artifacts,
out of frame, watermark, signature, deformed, ugly, mutilated,
disfigured, text, extra limbs, face cut, head cut, extra fingers,
extra arms, poorly drawn face, mutation, bad proportions, cropped head,
malformed limbs, mutated hands, fused fingers, long neck, logos

CFG Scale: 3.5
Sampler: Euler a
Steps: 24
Seed: N/A
VAE: N/A

Bigger Trash Cans

My preferred method of creating is to try and cultivate a prompt capable of generating interesting results and then trying to generate a series of images with that prompt. I know some creators prefer to do small batches or images or release only one at a time. I remember a photographer friend once telling me the different between the amateur photographer and the professional photographer was that the professional had a larger garbage can insinuating that the professional might need to take hundreds of shots to get one good one while the amateur can sometimes hyper focus on trying to take one perfect photo.

Advanced Settings

For these reasons I tend to set my CFG Scale to 3.5 allowing the AI more creativity. If I like the look and feel of the images, but the Checkpoint is deviating too far from what I'm looking for I might go up to 7. As for Sampler I often find Euler A generates the best results while being more cost (buzz) effective. If I'm trying to generate a better final image I might switch to DPM++ 2M Karras, but with a lot of Checkpoints I don't feel like I can tell much of a difference between it and Euler A to justify the higher cost. Steps I'll usually keep at 24 and I find that a good level to get a good idea regarding the quality of the output.

Buzz Cost

With these settings a single Illustrious image should cost 5 buzz (10% Creator Tip & 0% Civitai Tip) and two images cost 8 buzz. The cost will increase the more LoRAs you're using as well. I often will choose to generate two images because it will tell me a lot if the two images comeback very similar or wildly different. Does Lulu have the exact same pose and hand positions in both images or does it truly appear like dynamic pose or dynamic angle will have any effect. Based on those two initial images I will often test my initial prompt and try to decide which checkpoint I'm going to use for a post. Sometimes I even settle on more than one.

Hi-Res Fixing

While 2 buzz isn't much, it does add up over time and the added benefit is that when I scan my Queue I can more easily pick out images I've selected for Hi-Res Fixing versus the initial image generation. I will often set Denoise to 0.35 since I've found that this level tends to improve the image without overly distorting the character's pose, proportions or making major changes to the image. For roughly 9 buzz you can hopefully notice that Lulu's eyes look slightly more well defined, her hands and fingers are slightly more detailed and her outfit looks more detailed and in focus. There generally is no point in generating multiple Hi-Res fixes at 0.35 because in the end the images often turn out near identical

Bonehead Error: Be sure to check your Denoise level as well as your quantity of images. These value will often default to the last image settings you used and so pay close attention so you don't suddenly waste a lot of buzz unnecessarily. Measure twice, ~~cut~~ generate once.

Onto the Show

For this article I'll be posting two images from each Illustrious Checkpoint without any Hi-Res fixes or adjustments. I'll give my personal rating of ⭐, ⭐⭐ or ⭐⭐⭐ based on the results and how likely would I use that model based simply off the two images produced. If I encounter a two star model that I'm particularly interested in exploring further I'll add a plus sign (i.e. ⭐⭐+) Where applicable I'll add in any personal thoughts or opinions on the Checkpoint.

01: WAI-NSFW-illustrious-SDXL ⭐⭐⭐

The major reason I will tend to use WAI-NSFW-illustrious-SDXL is when I've combined several LoRAs together while using another Checkpoint and the image is just not working out. Often I've found that if I like the general content my prompt is producing but the faces don't look right or body proportions, I'll bring it over to WAI-NSFW-illustrious-SDXL and sometimes it just works.

02: Hassaku XL (Illustrious) ⭐⭐

A bit of a crap shoot, I've used it before in some of my generations, but for the most part I feel it can add embellishments that I wasn't asking for which can sometimes be a welcomed surprise, but frustrating as well.

03: Illustrious-XL - v0.1 ⭐

A bit too inconsistent for me usually while also being underwhelming more often then not.

04: Prefect illustrious XL - v3 ⭐⭐+

I have not used it a lot in the past, but it does have the 'look' I tend to lean towards when creating Lulu images or assets for stories.

05: Plant Milk 🌿 - Model Suite - Walnut ⭐⭐⭐

The Plant Milk model has become one of my three 'go to' checkpoints that I will start with. I tend to like that Lulu does not come out looking too anime while the overall generations that the Checkpoint produces will often be higher quality on first generations I've found.

06: JANKU v5 NSFW Trained + NoobAI + RouWei Illustrious XL ⭐⭐

07: Nova Anime XL - IL v11.0 ⭐⭐+

08: iLustMix - v9 ⭐⭐⭐

The iLustMix model has become one of my three 'go to' checkpoints that I will start with. It also does a great job generating illustrations while being flexible enough to generate some great NSFW content.

09: Nova Furry XL ⭐⭐+

I don't have an issue with Furries, but they are not generally what I'm looking to create. I've had some problems in the past with having to load down lots of negative prompts to avoid cat ears or tails. That being said I do like the consistency and poses of these test images.

10: Illustrij - v18 ⭐⭐+

I really do want to LOVE this model. The thumbnails they use always grab my attention, but I admittedly never seem able to produce similar quality work. I have not given up on it and if I'm casting a wide net looking for a Checkpoint to use on a project, Illustrij is almost always on my list to test out.

11: One obsession - 17_RED ⭐

12: Unholy Desire Mix - Sinister Aesthetic (Illustrious) ⭐⭐⭐

Unholy Desire Mix is the final of my top three most frequently used Checkpoints. There is just something about the way the images turn out that I think play a big part in the 'signature' Lulu appearance. It gives a nice illustrated appearance without being too influenced by anime styles.

13: CyberIllustrious | CyberRealistic ⭐⭐

Before I joined Civitai, Lulu originally started out as a 'realistic character' more out of necessity. I've not frequently tried doing realistic images of her, but it seems that CyberIllustrious does a good base job compared to others I've seen.

14: Illustrious Realism by klaabu ⭐⭐

Same sentiment as I expressed for CyberIllustrious. I've not played around much with realistic Illustrious checkpoints, but I like the results from this test.

15: Citron Anime Treasure [Illustrious & NoobAI] ⭐

16: Nova Orange XL - RE:v2.0 ⭐

17: Hoseki - LustrousMix [IllustriousXL/NoobAI] ⭐⭐++

This is the first time I think I've generated anything with Hoseki and I'm really pleased to discover it. This one will definitely be one I want to check out. Enough so that I'm giving it and extra '+'.