🌐 Online generation available at Shakker.ai & SeaArt
✨ VERSION 2.0 · REWORK
Version 2.0 is a complete rework. This version addresses some issues from previous versions, offering cleaner image quality, enhanced overall realism, and noticeable improvements in facial features.
However, the use of Adetailer or Face Fix is still highly recommended.
🏠 Recommended Parameters for Local Generation
Samplers:
Recommended:
DPM++ SDE
DPM++ 2M SDE
DPM++ 2S a
DPM2 a
DPM++ 3M *
Euler A
Settings:
Steps: 25-35
CFG: 4-5
Clip Skip: 1
Resolution: >1024
Additional Information:
Euler A provides a soft and cleaner appearance, while SDE samplers produce a more detailed and sharper look. Experiment with different samplers to achieve your desired look
Use Danbooru tags.
Prompt Style:
Positive:
masterpiece, best quality, amazing quality, realisticNegative:
3d, anime, 2d, bad quality, worst quality, worst detail, sketch, censor, simple background
🏋️♀️ Training LoRas
Read the following article for tips and my training preset
🌞 Final Remarks
This will be an ongoing development aimed at improving the realism and various aspects of the model, so feel free to share your feedback.
Thank you for downloading and using the model. I hope you enjoy it!
🪄 Workflow
⚡️ Buzz for the Best Images ⚡️
Description
FAQ
Comments (19)
Test images: https://civitai.com/posts/9297197
Rillusm v1: https://civitai.com/models/959736/rillusm-realistic-illustrious
Big Love v1: https://civitai.com/models/897413?modelVersionId=1004186
Rillusm v1 with Big Love v1 as refiner at various starting points.
This combination gives you the creativity of Illustrious with the photorealism of SDXL.
Big Love v1 is a Pony based model that has extreme realism but lacks flexibility.
(Images created with both checkpoints using Hyper LoRA, at CFG of 2, steps of 8 and sampler: Eular A SGM Uniform. Hi-res of 1280x1856 at 40% denoise used. Drawthings app for Mac. 70% is the same as 0.70 denoise)
Image 1: Rillusm v1 only. Refiner not used.
Image 2: Refiner starts at 70%
Image 3: Refiner starts at 50%
Image 4: Refiner starts at 35%
Use a normal SDXL model as refiner and the results will be cleaner
Hahah! WTF!
Firstborn again.
Anyway how will be this better, than realistic pony models?
I see, it has better lightning and background already.
Push it toward. Good job.
I'll just say, it makes really good dummy thicc Xenomorphs! Well done 👽
yeah I saw your pics and it looks really crazy... well done haha
@Tetsuoo Why thank you!
Realistic Illustrious changes my life!
Already love it, can't wait to see how it evolves!
❤️❤️❤️
what exactly changed your life?
first thoughts - there is a strong Asian bias for faces, and skin texture is very plastic. However this is a huge step towards realism! I really love it and I'm having a lot of fun trying it out.
Yeah, its really strongly trained on anime, so it will take some time
@ZyloO absolutely, thanks for the hard work
Who needs FLUX now? Superb prompt adherence!
As someone mentioned, a bit of an Asian bias, though not a big issue. And males sometimes look a bit female-ish, but this is inherited from the base model.
This is still with tagging styled prompts, you cannot compare it with natural language. Nonetheless, the image quality is superb.
For first try, its good. Its semireal and has asian bias, because of base model.
It just needs several thousand new data, and will be good.
Nice first step. +1 to observation of Asian bias. More training and this will be fantastic checkpoint.
How big was the training dataset for this?
does it have vpred?
The model is pretty awesome and has potential to be as realistic as possible, for now is a good 3D one but a good start
WOW, keep up the good work, Thank you!
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.