🔞 A super-realistic model created with the purpose of creating as realistic girl characters as possible. Combining a lot of different strengths from 11 different base models and manual adjustment for the most realistic result.
✅ The model has been tested with thousands of images and different LORAs. All the current LORAs are suitable for use with the model, and adjusting weights if needed to get the best result.
🔔 This 3.0 version will even push the "realistic" element further. Enhancing facial and eye details. Please make sure to install AFTER DETAILER for the best result.
Suggesting sampling method: DPM++ SDE
Suggesting sampling steps: 8-16
Suggesting CFG scale: 2-5
All the monetized purposes using this model without my commission are FORBIDDEN.
Please CREDIT me if using this model on any work, non-commercial project, or images.
Contact me at: [email protected] for any commercial use of this model.
Description
UnrealWorld Model v3.0
A highly realistic AI model made to create lifelike female characters. This version is built by combining 11 strong base models and manually adjusted for the best and most natural results.
💡 Features:
Super Realistic: Improved face and eye details in version 3.0 for even more realism.
LORA Friendly: Tested with thousands of images and works well with most LORAs. You can adjust the LORA weights for better results.
Celebrity-Ready: Can create famous female celebrities with the right LORA and settings.
Flexible Style: Great for both soft natural looks or sharper, more detailed images.
⚙️ Recommended Settings:
Sampling Method:
DPM++ SDESteps:
8–16CFG Scale:
2–5Use AFTER DETAILER for best results
🚫 Rules:
Do NOT use this model for paid or commercial work unless you get my permission.
Always credit me when sharing images made with this model (non-commercial use is allowed).
📩 For commercial use or questions, contact me at: [email protected]
FAQ
Comments (13)
This is Incredible!! Thank you very much.
Don't know why, the model seems to be working strangely with SD.next and Zluda. The image is like "oversaturated/burnt" even with a guidance of 1.
See links, the 1st image is done with my merge, the 2nd and 3rd (guidance 1) with this model:
https://files.catbox.moe/rk2xi6.jpg
https://files.catbox.moe/dl0lxt.jpg
https://files.catbox.moe/ypnqhv.jpg
Never had this before.
EDIT: I of course tried your negative/settings. Doesn't change much things.
Can you please provide the prompt so that I can look into further details. With only the first look I'm assuming you are using too much "weight" on your lora and there might me some negative embeddings that effect the quality.
That's not the lora, I ran tests without any lora, same thing.
I don't use any embeddings, in positive or negative prompts. Like I said, I tried the exact same prompt/setting/cfg than one of your pics and had the same sort of weird result. I'll reproduce one of your pic and I'll edit this post with the pic attached.
(And the size of the model is a bit strange too. EDIT: cause of duchaitenaiart_V453)
The metadata are embedded in image files. But I reproduced this pic you posted:
"photo of 18 y.o cute latina girl, dark skin, ponytail hair, etc. etc."
https://files.catbox.moe/7m8a70.jpg
https://files.catbox.moe/itsbos.jpg
Since I'm using SD 1,5 running with Zluda, maybe one of the checkpoint you used don't like it. I'm a bit lazy trying every model you merged one by one though. :D :D :D
EDIT: I'm trying all of them, anyway....
epicphotogasm_x: Checked. Works "normally"
@NxtAI Found the culprit:
realisticVisionV60B1_v51HyperVAE
That's the hyperVae merged in it.
Pic: https://files.catbox.moe/yw12el.jpg
@charless Thank you for bringing this to my attention. I hadn't noticed these issues before, but I'll address them in the next update and continue improving the model.
@NxtAI I could guess it's not a big issue on "regular systems" running on NVIDIA, maybe just for users using Radeon cards :)
Nice sleuthing charless - I'm having the same issue. A1111 + Zulda + Radeon card...
For note - tried turning off using a local VAE but it still has the same issue.
@WhoAmI112233 yeah unfortunetly if it's merged into the model, there's nothing we can do. :)
great model, having fun with it!
Pruned version?
good trained model, giving me a lot of nice results. 👍
is there inpaint feature?
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.