Information:
This model was merged and trained on women only. It can depict males but is not very flexible.
LoRA Training:
Use either wdv3 captions or real captions, i personally use Florence-2 Large in TagGUI
Train on base pony, results are close enough imho
Use Prodigy Optimizer with a d-coeff of 0.5
Use masked training with an 1.0 alpha mask for the face and 0.3 for the body, if possible remove hands and feet from the mask
Do not train the text-encoders, only unet
600 - 1000 steps is suggested
Suggested Settings:
Sampler: Euler a + AYS (Align your steps)
Resolution: 832x1216
Steps: 14
CFG: 4 - 7
Perturbed Attention Guidance 2.7
HiRes Upscaler: 4x_NMKD-Siax_200k
Sampler: DPM++ 2M SDE + SGM Uniform
Steps: 30 - 50 (Or 12 with SDXL Hyper LoRA at 0.5 weight)
Denoising Strength: 0.25 - 0.35
Prompt:
As usual start with the score tags. I personally start every prompt with this huge list of tags:
score_9, score_8_up, score_7_up, photo, source_photo, source_real, RAW, 8k, low quality, traditional media, amateur, real life photo, film grain, dark, darkness, dim night light, light flash, very dark focused flash photo, (dark background:0.75), low light, detailed faceDatAss Rev 2 Recipe:
Step1: GameArt Pony + CyberRealistic Pony (0,0.1,0.2,0.3,0.3,0.3,0.3,0.3,0.3,0.2,0.2,0.2,0.3,0.3,0.3,0.3,0.3,0.3,0.2,0.1)
Step2: Step1 + Custom Trained Pony Model (0,0.15,0.25,0.3,0.25,0.2,0.15,0.15,0.1,0.1,0,0.1,0.1,0.15,0.15,0.2,0.25,0.3,0.25,0.15)
Step3: Step2 + (Damn Pony Realism - CyberRealistic) (0,0.35,0.6,0.65,0.45,0.25,0.1,0.1,0.05,0.05,0,0.05,0.05,0.1,0.1,0.25,0.45,0.65,0.6,0.35)
Step4: Step3 + (Personal Anteros Mix - SDXL) (0,0.25,0.4,0.45,0.25,0.15,0.1,0,0,0,0,0,0.05,0,0.1,0.15,0.25,0.45,0.4,0.25)
Step5: Step4 + (Step4 (+SG Finetune) - Step4) (0,0.2,0.55,0.43,0.35,0.1,0.05,0.05,0.05,0.05,0.05,0.05,0.05,0,0.05,0.05,0.05,0.05,0.05,0.05,0.05,0.1,0.35,0.43,0.55,0.2)
Step6: Step5 + Personal Anteros Mix (0,0.1,0.28,0.28,0.2,0.1,0,0,0,0,0,0,0,0,0,0.1,0.2,0.28,0.28,0.1)
Step 7: Step6 + ICKPOTLycoXL 0.25 + photo_v2 0.2Description
Trained a bit more, then added a lot more checkpoints and LoRAs to the mix, improved contrast and realism.
Not much room for improvement here, it might not be better than Rev 2 in all aspects.
FAQ
Comments (17)
For sure the best model for realism, hands down. Insane job!
It's the most coherent realistic Pony model by far, but it doesn't look very artistic and creative sadly.
@virtualfix6885 in terms of photorealistic NSFW pony models, what would your top 3 be? I've been bouncing around a couple now, and there's just too many. Before I knew it, my google drive is full
@BobTheBurner Yours, Zonkey, DAMN. Those are the best realistic models, with DAMN being the most creative.
If you would have told me that irl, princess zelda is such a naughty girl... Good job :)
So far this is my favorite realistic model for generating women. The quality is great, it still remembers most Pony concepts, and you don't need a wall of schizo negative/quality prompts for it to work properly.
I wish it could generate better looking dicks in POV pictures though lol.
Excellent... do you have also a configuration to train a finetuned custom model on top of your checkpoint? I prefer first to do finetune, then lora extraction...
hi, tnx for this, could u please mix the 2 step lighting lora to and make a new checkpoint it so us cpu users can generate faster? thank u
Finetuned a person with this model and turned out really good. Perhaps even the best of the ones I've tried so far. Still testing but so far it looks like a great model
Edit: One negative thing is that the face and especially eyes are often blurry
I think I tested about 10 pony realism models and so far this one looks the most realistic to me. Really great job. If I had to nitpick, I would say that some faces are not as beautiful (aestethically that is) as with some other models. Also, I feel like some faces are reappearing a bit, but I am not sure. But this is only meant as constructive feedback. This is my favorite model so far and I played around with it a lot already.
when training, does your suggested learning rate apply to unet learning rate too?
Probably the most realistic looking pony model I've tried, it does seem to really want to only give women three fingers per hand though
Yes this model is particularly bad at hands unfortunately.
I wanted to express my gratitude for your amazing model. It does a fantastic job at interpreting realistic pony characters. I have been using it extensively and truly appreciate your contribution to the community.
However, I have noticed a recurring issue in my tests. The model tends to generate backgrounds in a beige or cream color, regardless of the prompts I provide. I believe this might be an area that could benefit from some adjustments.
Thank you again for your excellent work and for considering this suggestion. Your model has been incredibly helpful, and I look forward to seeing any improvements you might make.
One of the best realistic models! Outperforms other realistic ponies much of the time, particularly for realism and skin texture.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.