Come hang out, test LoRAs 🧪, give feedback 🗣️, see WIPs 🎨, and more. 👉 Join here: https://discord.gg/WeqYRZudhW
Illustrious Realism by klaabu
For on-site generation use -
Dpm++ 2M Karras
CFG:5-6.5
Steps 25-35
sdxl_vae.safetensor
On-site V1.0 now has VAE baked in
For local generation use -
DPM++ SDE
CFG: 3-4.5
Steps 25-35
sdxl_vae.safetensor
A cinematic fine-tuned model for high-detail, character-rich realism
"Illustrious Realism by klaabu" is a high-fidelity model fine-tuned on top of the Illustrious framework, enhanced for realism, texture depth, and stronger control over nuanced facial and body features. It has been tuned to preserve the cinematic essence of Illustrious while elevating clarity and dynamic lighting.
Native Resolution Range:
Outputs cleanly from 832×1216 up to 1536×2048, ideal for both portraits and full-body shots.
Recommended Inference Settings:
Sampler: DPM++ SDE KarrasCFG: 4.0Steps: 20Clip Skip: 1Prompting Style:
For best results, use Illustrious-style prompting — avoid stylised scoring systems like Pony. The model performs best with clean, natural phrasing and descriptive realism.
Suggested Positive Tags:
masterpiece, best quality, ultra-detailed, realistic skin, intricate details, highres,Recommended Negatives:
low quality, worst quality, blurry, (deformed:1.3), extra fingers, cartoon, 3d, anime, bad anatomyCharacter Handling:
The model handles a wide range of character likenesses including many known to Illustrious. It works smoothly with compatible LoRAs — including SDXL LoRAs, Illustrious variants, and Pony.
Illustrious Realism is crafted for users seeking high-end realism with expressive tone, dynamic lighting, and compatibility with advanced prompting techniques — all within a flexible resolution range.
Please support my work
If you like what you see, share your generations and give the model some buzz. Seeing your images helps improve future versions and keeps the project going.
Description
For on-site generation use -
Dpm++ 2M Karras
CFG:5-6.5
Steps 25-35
sdxl_vae.safetensor
FAQ
Comments (45)
I'm eager to try this out tomorrow! Your Amateur Slider in my couple attempts didn't achieve what I was personally quite looking for (maybe bad usage or unfavorable checkpoint combo) but seeing as this is Illustrious! Because Cyberdelia's Illustrious + Zyloos Amatuer Lora is the combination that I've seen that offered me the most realistic realistic photo/natural lighting to date. promise to try it asap
EDIT: What is Illustrious-style prompting, I don't know it, I don't even necessarily know pony prompting either. (aside from scores) I usually just put a longish main subject then misc danboru tags.
Any examples of male-focused images?
https://civitai.com/posts/14910391
Once civit allows to upload images to this checkpoint, I'll update. Thanks
beautiful!
🌟🌟🌟🌟🌟
Must have model
i can't generate high quality images with this template and i don't know why. I'm generating them using tensor art, using the same steps, seed, positive and negative prompt, etc. Is there something that's not in the description of the images that I need to fix or is it just the site? this is the best realistic model based on illustrious that I've found and I wish I could use it better
tensor art uses their own TAMS generator and civitai uses comfy
also civitai has mandatory negative embedding you can't disable
@Nyaa314 well...shit
@Leamas_ this model shouldn't be available on tensorArt. Can you DM me the link?
@klaabu tensor art allows you to upload any model you like, even if it's not yours, provided you make it clear that it's a reprint and link to the original model/creator. In the case of this model, I uploaded it because I wanted to use it and my computer can't handle xl or heavier models. If you want, I can delete the model from tensor art or you can cleimar it (something intended for the original author, I put it as a reprint).
https://tensor.art/models/847012935961255784
Can you tell me which VAE you are using
Sdxl_vae, other settings as per example or have a look at the model info :)
Had really great hopes as I liked the realism it generates, but too bad just like other pony anime garbage, it cannot generate dark skin characters.
Hi @b3lli3 , I appreciate your feedback. I believe you must be prompting something wrong, if you have a look at the pinned post, you can see how easy it is to create different ethnicities like - African, Asian, Indian, Mexican etc. Ensure you read the model information if genning on site, if genning locally, use the settings from cover photos - DPM ++ SDE, CFG 4, Steps 25+ .
PS! This is not a Pony model, do not prompt it as such. Ensure your prompt is for ILXL.
Good Luck!
@klaabu Thank you, sorry if my feedback sounded harsh. But sometimes I become tired as it is quite difficult to get the different skin textures other than pale. I will try again but maybe on comfyui with the settings you recommended, thank you.
@b3lli3 no problem. Not harsh at all, I' here to help. Take one of the cover images or the example images of different ethnicities and work on that. While generating dark female ans white male is hard on most every checkpoint, it's not impossible with right prompts ans negatives. Let me know if you need hand with anything else :) take care
Why I have dark skin babies just fine?
Some "photorealistic" mixes indeed have only white skin, but it is super rare case.
Absolutely can gen any color skin. This is a really great base model. Sweet spot for me has been SDXL 896x1152, DPM3++SDE, Karras, 50 steps, 6 CFG.
This model responds VERY well to higher base resolutions. In fact, my gen’s are absolutely garbage anything below 768x768.
Try using weights like (brown skin:2), (black skin:2), etc.
@primez19 if you gen locally, try 2048 x 1536 - it will generate well at high resolution without upscaling :)
Does anyone know why these blue dot anomalies are happening in the images? I don't know what to do.
@d4nt3gm136 , if you read model Info, it clearly outlines what settings to use. For on site generation used dpm ++ 2m karras with sdxl vae. CFG 4-6 and steps 25-35.
Its VAE, try another one.
@klaabu You seems have no idea what they talking about, blue dots has no relation to settings at all...
@littlefluffyball thank you for this great observation.
@littlefluffyball Which one do you recommend? I want to create realistic skin AI models
@d4nt3gm136 , it's written in model Info - sdxl_vae.
@klaabu thanks
Your checkpoint file named:
illustriousRealismBy_v10.safetensors
When automatic downloaders like aria2 sees that same file already exist, it will not download.
Hi @littlefluffyball , Ah, got it! Appreciate the heads-up—but just to clarify, the file was uploaded with the full name: Illustrious_Realism_by_Klaabu_v1.0.safetensors. Looks like CivitAI renamed it on their end, which is pretty common.
As for aria2, it’s doing exactly what it’s supposed to—detecting a duplicate file and skipping the download. That’s... kind of the whole point of it, right?
Would you like me to personally rename the file for you next time too? Just let me know—happy to help streamline your download experience.
Hello, I've tried to use this checkpoint on CivitAI, and it is interesting in the quality of real people look, it can give.
But the fact it doesnt have the VAE inside is a big problem. If you give "REMIX" on other's images, the generator will not charge the VAE and you will have washed results. I think 90% of people leave after this. But reading your info I managed to manually load the VAE and the images are perfect.. only to discover it will not work with hiresfix... So you cant use it on CivitAI generator!
In the end, my suggestion is to put the VAE inside the checkpoint, like everyone else does...
@settima_ai , thank you for the feedback, I really appreciate it. I've noticed this myself, but leaving VAE our was intentional. Civit is quite limited when it comes to on-site generating and there are better options available off-line. While I agree, I should have anticipated the fact that people don't read info and generate washed out images. A new version with VAE baked in will be available in next week rotation.
Hires-fix is Civits problem again, it's a known issue and nothing I can do about that.
I appreciate the support, and stay tuned for VAE release. If you can, please run this model locally as it's potential is way beyond what Civit can generate.
❤️
Great work on the realism! Usually, I find and try new checkpoint models within the on-site generator since they show you a listing of other Illustrious or Pony models and I just find the realistic ones there. I don't read the model page until after the first generation just to see what it will do. If the image looks washed out, I know that it needs the VAE, but probably most people don't realize this. I look forward to the baked VAE version.
@settima_ai @ponyboi civit has fixed the sdxl and hires fix bug. Happy genning.
good but seems VERY prone to long torsos. everything else is fantastic
Hi @vengeancedim47317 , thank you for the feedback. Noticed this too when genning in native and above and up to 1536×2048. Either add "long torso" or follow aspects rario from 832 x 1216 up to 1536×2048.
❤️
Can't find the sdxl_vae to add to not get those blue dots so sadly I cant use it. Maybe try to bake it into the model next time?
Still getting used to using it instead of Pony, but getting very consistent and realistic gens now. Also seems to handle faces, fingers and feet a bit more reliably. Found it particularly good for Pinups and vintage amateur stuff. Love it ❤️
This is capable of creatin very realistic images, worx well with many Loras, overall a really really good checkpoint 2 use for super imagez. Thnx 4 creatin it!!! Makes Civi use more fun 4 sure!
A+! My new favorite Illustrious model! Just tried a bunch of realistic models and this one has the best photographic compositions, the least amount of bad anatomy and looks the sharpest all while producing highly believable realism. It's well suited for professional studio (w/ Evil Angel LoRA) and amateur NSFW photography. Thanks for making. I've barely scratched the surface of this model, but I'll be sure to provide feedback if I find anything that needs work.
@ponyboi , thank you so much for the feedback! Please ensure you prompt it like you would on illustrious model, it does understand natural language, but to get the best out of it - ILXL prompt. It also knows same characters etc like ILXL base, although sometimes struggles making certain characters without proper prompt guidance. ❤️
Amazing results between reality and fantasy but somehow I just can't generate futa models
Hi @challasvarias01106 , ensure you don't prompt like you do on Pony. Needs to be prompted like you do on Illustrious - futa, erection, (futanari with penis), large testices, . Works fine
Add to negative:
(Long Torso),(Elongated Torso),(Stretched Torso),(Overly Long Body),(Bad Anatomy),(Wrong Anatomy),(Too Long Torso),
Adjust weights / add remove tags as needed. Also, it does help to use LoRAs that are made for anatomical purposes. Like PornoMaster's Skin Texture LoRA for Illust which is VERY good and doesn't ruin the style of things too much. Obviously adjust the weight as needed.
It should be noted that you CAN go above 1.0 with Illustrious / Pony Loras. That old "Don't go over 1.0" thing is no longer applicable and going higher can yield better results.
But yes, watch your weights and how you combine things.