
You can get this model, along with many others, by joining The Tinkerer on Whop. It’s one monthly membership, and you’ll also get early releases, private tools, and a bunch of extra stuff.
CyberIllustrious blends the artistic flair of Illustrious-XL with a big dose of CyberRealistic’s style. You still get all those cinematic, dreamy strokes, but now they come through with crisp, realistic detail.
If you want that painterly, elegant look but still want your images to feel sharp and lifelike, this is your model. It’s great for editorial vibes, soft glam, or even fantasy art that looks like a real photo. Basically, it just nails it.
It’s powerful, easy to use, and honestly? Prompting it is seriously fun.
📘 Get the most out of this model with the Illustrious / NoobAI Prompting: Master Guide.
⚙️ Recommended Settings
Sampling method: DPM++ SDE Karras / DPM++ 2M Karras - SGM Uniform / Euler a
VAE: is already Baked In
Sampling steps: 30+ Steps
Resolution: 896x1152 / 832x1216
CFG: 4-5About (DMD2)
The special DMD2 edition of CyberIllustrious is created for faster, cleaner generations while requiring fewer steps.
DMD2 takes the balanced improvements of the standard release and optimizes them for speed, precision, and responsiveness, making it ideal for both SFW and NSFW workflows.
Expect tighter results, smoother control, and a model that reacts beautifully to the classic Illustrious prompting style.
Recommended Settings
Sampler: LCM
Schedule Type: Exponential
Sampling Steps: 8 (12 for extra fine detail)
CFG Scale: 1.5
⚡ These settings are optimized for DMD2 and are different from my standard Illustrious workflow!
🧠 Prompting Made Easy
Positive Prompt:
masterpiece, best quality, high quality, ultra-detailed, realistic, ... rest of promptNegative Prompt examples (optional):
cartoon, illustration, anime, painting, CGI, 3D render, low quality, watermark, logo, labeldeformed, deformed face, low quality, bad quality, worst quality, (drawn, furry, illustration, cartoon, anime, comic:1.5), 3d, cgi, extra fingers, deformed teeth, watermark, logo🩺 Optional ADetailer Settings
Adetailer model: face_yolov9c.pt
If you only want the main face being refined set 'Mask only the top k largest' to 1.🔗 Resources
Backup location: HuggingFace
☕ Support the Madness
Liking what you see? Grab me a digital espresso (or a couple!). It keeps the creative energy going and helps bring more cool prompts and models your way.
⚠️ Heads Up
This model can spit out some mature stuff. Use it wisely, and please, no cursed creations—we’ve really seen enough of those.
💬 Want Better Prompts?
Use this custom ChatGPT to auto-generate polished prompts for CyberIllustrious:
🔗 [Try it now on ChatGPT]
Description
Since version 1.5, I have been working on this version. It is based on a different internal foundation, and I will also use the next versions on this foundation. Minor issues I had have been resolved in this version, and it is a more stable version that is very user-friendly and very robust.
Yeah, yeah, I know - this version dropped faster than a new Marvel trailer. I've already got my popcorn ready to watch the reactions roll in. 🍿😆
FAQ
Comments (63)
This new model saved my cousins marriage! thank you!!
Wow, it sounds like your family has a real knack for getting into marital pickles!
so whats the key differences with 3.0 and 2.0? i loved using 2.0 already so i'm keen to know if there are any differences aside from the foundation change you listed :D
The result is different. Normally, you get the same type of image when you use the same seed, but that's not the case now. In some cases, V3 is more realistic than V2, and you get fewer strange body parts, etc. V2 was/is also fine, but V3 is probably now the basis I will continue with.
I could have chosen not to publish it and wait until it is further improved - but my idea is that it might be good for others as a basis/part for their checkpoints.
@Cyberdelia so overall more realistic and fewer errors on bodies, sounds like a W to me :D
Really fun model, getting really creative results from the same prompt
Looks great. More men, pls :)
3.0 is amazing, really creative results thanks for sharing
Love your images (V2). Excellent stuff!
Just started with version 2 and there's 3 already. Works great so far.
Can you develop a realistic model based on the noob eps model? Its character concepts are very rich. Almost all the characters in the danbooru website with more than 100 pictures can be reproduced, eliminating the need for most lora.
In addition, although the 3.0 model is still very good, it has begun to lose some role concepts compared to the 2.0 model.
How do you handle the gazillion character without loras?
I use lora for characters, just because i can save them with pics and prompts. So when i want to use them, i just click on one lora.
@ellistoby621 When there are enough things to do, you need to use a lot of LoRa. In this case, the LoRa will conflict with each other, so a good model should save the use of LoRa as much as possible. For example, in most cases, the accuracy of characters and clothing does not need to be too high.
@Cyberdelia do you plan to release a FLUX Cyberrealistic checkpoint? Been using your models since a long time, but recently focusing on FLUX.
Thank you ultra Cyberdelia! In your personal ranking, how would you sort your models purely based on the reality score?
I would love to be able to generate with this model onsite
This checkpoint legit just blew all illustrious checkpoints out the water. Prompt adherence, clean generations, and the best adherence to ethnicities.... ty for this!
Pretty impressed by this checkpoint, I'll make some lora/lycoris models for it. ;)
There's no chance I can generate futanari, I tried v2 and v3 and they are just picture of sex with a man.
Any advice?
this depends on what version CyberIllustrious you are using, the tags would depend on what AI service you use too, (A1111, ComfyUI, etc), but if using A1111, try using the sd-tagcomplete extension then use prompts such as (1girl is a futanari, penis:1.4), (full-package futanari:1.4), penis size is up to you and same for whether you want testicles or not, hope this helps :D
@NightKillers101 Thanks for the response, I'm using ComfyUI. I don't know if any other extension would help. But I will try these prompts that you mention.
@challasvarias01106 since you're using Comfy, the prompts will likely help you out and i'm unsure on the extensions, i have tried Comfy before but never figured out how to use it, so i stick with A1111 personally.
Good shit
This is a very good model. The detail density is higher than other XL models. I almost never had photos with bad hands and I am very surprised by this). The colors are rich, the white haze adds realism. It seems to me that this model works better with lora than other models. The model also generates a variety of photos, which speaks of the quality of the content. I actually got cool photos with this model. Thanks to the author!
Hi Cyberdelia.
If Pony7 will be good, (i hope it) will you jump on it?
I think, the big deal will be the out of box realism. With finetune, even better.
I wont jump on Illustrious till then. Its now clearly better, than pony finetunes, but im patient.
I wrote thise as a review but not sure if its easily visible so writing here aswell
I've just started playing with it yesterday and will somemore in the following days. review so far
+ Great adeherience
+ Awesome ability to produce very realisitc natural photos. amatuer aestheitc with natural lighting.
- Usings tags such as Big/huge for bodyparts can result in thicker/bigger overal bodies unwantedly.
- As a double edged sword to the above positive, Though you can achieve a amatuer aesthetic, even when , images can exhibit , those images will have general low quality even after using upscaling techniques within GUIs. Such as course/rough textures and blurryness.
V3 is based on 1.0 Illustrious? No fuckin lora works.
Characters.
Which models are you trying? IL lora models work fine for me.
Can somebody confirm, that v3 is incompatible with characters? Even if i prompt the name. What happened?
Thats a shame, because Cyber models were always for compability.
This is a fundamentals question: What is the difference between Illustrious and Pony?
For anime: built in styles and artist, much more characters, more usable booru tags (newer dataset). Better adherence, better lighting, better backgrounds.
For realism: Same as anime, more realistic (with good prompt). Sex poses must be prompted by strict booru tags, otherwise you can use more natural language.
very good model thanks to the author
Version 3 is an excellent model see the rendering of the images
While Shitlustrious shitting on users with incompatibily, you should just drop this.
Focus on Noob vpred, which surpass the unreleased Illus models too. It wont be change in the near future, thatswhy the best base model to train on.
Tried to compare this and Cyberrealistic Pony and this is better almost in everything, for me. But in Pony there's more realistic bodies and skin texture, also it seems that it understands body tags better. But overall understanding and variety is much better in Illustrious. Also barely any lora seem to work.
You must use strict booru for sex
https://danbooru.donmai.us/wiki_pages/tag_group%3Asex_acts
And yes, 0.1 loras barely work.
@jasonjacoby474 not talking about sex acts, just the body composition and skin texture. But thanks for the link, useful info there.
It is insanely good, like all possible Loras packed into one checkpoint, also it understands prompts almost perfect. Anatomy and faces are great.
Only minor drawback found - seems like it cannot do art styles like pastel, watercolor, oil painting etc. But it is not a big deal.
Just, WOW! 😍
I found my way here because every time I did a filter I would find Illustrious content popup and then I noticed oh it is a new checkpoint type. Next logical thing was see if there was a cyber version..bingo there is I love your other stuff, 1.5, pony real and semi.
Question: Do you find this is better/easier to prompt? Gets better or similar results as far as anatomy, accuracy of details etc? Is it really only for people or like the other checkpoints you do can do other interesting things as well?
IMO, compared to pony and some SDXL models I tried, prompting is much easier here.
This is the best realistic Pony Checkpoint I've ever used
what kind of Loras work on Illustrious?
Best part of this site is the filter option. Just go to Models from main bar, then on the right click the filter option. Then tick off Lora for Model type, and Illustrious for Base Model. Then you will only see Loras that work with this Checkpoint.
@Oddkward oh okay, I asked because I tried a pony Lora on it an I swear it works
@TheCAL Pony and SDXL LoRAs (and I think LyCORIS, DoRA also) technically work with Illustrious base models, but you may need to adjust (often lower) the strengths to get more ideal results. Pony LoRAs seem to work a lot better for Illustrious base models than SDXL ones, from my experience. For NAI/NoobAI LoRAs you might also need to lower the CFG settings to avoid extreme over saturation.
@cmk548 Yeah, IMO XL and Pony LoRAs look better on Illustrious than on Pony. Illustrious ones look best of course, but it's all relative. Initially, I didn't know we could use Pony on Illustrious and it was a pleasant surprise when I tried it. Great backwards compatibility.
This model is very good in its generations but one huge problem is this check point doesn't recognise nationalities, the pony model has the same problem but the original but the original cyber realistic xl is very good nationality eg i try korean, thai, filipino or malay it's always white girls that come up, anyone found a work around?
Pony and Illustrious are anime-based engines so they hardly tuned towards one character type. Try to experiment with LORA: https://civitai.com/models/119376?modelVersionId=608750 or make inpainting using regular SDXL-based model.
Really good model, my favorite right now, doing well in bringing those details to life. It appears to be very limited when it comes to generating different ethnic groups other than causacians though.
Because its based on anime dataset.
Does anyone know a good method to get rid of elongated torsos that plague many illustrious models?
Portrait resolution(896x1152 or other)
(Long torso), (elongated torso),(stretched anatomy),(wrong anatomy),(bad anatomy),(overly long body proportions),
This works almost all the time for me. I change the weights for each one at times until things work. Also, using Loras that have certain styles or say skin texture modifications seem to work very well too. Often times I can get correct proportions.
It's a feature, not a bug ;)
--------
Generate pictures from all angles, then use antilongtorso.lora in the station to generate another short torso picture with the same parameters, and label the original picture with a tag, such as long torso, as the reg picture for training lora. After the training was completed, it was merged into the model, and long torso was added to the negative prompt word.
@suede2031691 idgaf about what is or isn't permitted, if the model isn't accurate I won't keep using it. I'm not trying dance around censorship just to get my good boy medal. either a model is good or isn't. but I wasn't criticizing this one in particular, I noticed it's a trend in illust models. Even when using negative prompts with large weights it just doesn't understand correct belly anatomy. Sexualizes everything, and then tries to invert the roles saying we shouldn't sexualize minors, rofl. Then train it so it doesn't?
What kind of upscale settings are you using?
I normally use img2img with the same settings as the original via pngInfo and 0.2 denoise with 2x upscale and remacri.
This works well for CyberRealisticPony and other pony models it often increases the final quality, i.e. fixing small things and add details. However, with this model, the best I can hope for is a same quality and it often introduces artifacts, such as splotchy skin, and sometimes the output is just broken.
i ran into the same thing. you can do extra passes with a much higher noise than XL models. I find that 0.6-0.7 noise with a 1.25-1.5 latent upscale works similar to a ~0.2 w pony. illustrious is definitely more temperamental, but this model works pretty well. Euler A / normal seems to be a little bit more stable but with less micro detail.
Details
Files
cyberrealistic_v30.safetensors
Mirrors
cyberillustrious_v30.safetensors
cyberillustrious_v30.safetensors
CyberIllustrious_V3.0_FP16.safetensors
cyberillustrious_v30.safetensors
cyberrealistic_v30.safetensors
cyberrealistic_v30.safetensors
cyberrealistic_v30.safetensors
cyberillustrious_v30.safetensors
cyberillustrious_v30.safetensors
cyberillustrious_v30.safetensors
cyberillustrious_v30.safetensors
Available On (3 platforms)
Same model published on other platforms. May have additional downloads or version variants.



















