If you like what I do please consider supporting me on Patreon.
Pony Lewdfusion is an extraordinary, groundbreaking checkpoint that pushes the limits of reality and imagination to bring forth the most hardcore, jaw-dropping, pornographic masterpieces ever known to mankind. With a proprietary blend of cutting-edge machine learning algorithms and neural networks, Lewdfusion harnesses the power of advanced deep learning techniques to generate images so lifelike and enticing that they rival even the most talented and experienced photographers in the adult entertainment industry.
VAE: Baked in
Sampler: DPM++ 3M SDE
Steps: 30
CFG: 4.5
Invite to my public discord: https://discord.gg/UFkPTfTDWs
Description
Largest leap forward in realism compared to any previous versions.
FAQ
Comments (12)
Looks promising
4 step lightning version please
5.0 is Great. I made some mistakes comparing it earlier & I'm glad I was wrong. It turned out great! I'm getting some of the most lifelike photos I've ever seen from a realistic pony model without having to flood the negative prompt section.
In fact, I just double checked and all I have in negative prompts is: score_6, score_5, score_4, censored.
The results are amazing. As we all know, there's normally a 'worst quality, poor quality, monochrome, ext.' thrown in there.
Very impressed!
Worth the buzz to anyone that's on the fence!
Thanks for the review that is great to hear!
Never been a fan of realistic models, since it ends up creeping me out, but man, this models is calling for my buzz... nice work!
This model is really good at creating photorealistic images while still preserving most PDXL features. Although v5.0 generally looks slightly better than v4.0, it lost some of its capabilities to create images of pretrained characters compared to v4.0.
Hey thanks for your comment! I'm curious about this, can you give me an example of one of the characters 4.0 could make but 5.0 could not? I will see if I can balance the two, but it might just be one of the trade-offs of pushing it in a more realistic direction.
@getphat It's not that v5.0 completely lost characters, but if you try to create the same character with v4.0 and v5.0, most of the time v4.0 gives me a more accurate image of the character. You can see this for example for Zelda in my test images:
https://civitai.com/images/33480380
https://civitai.com/images/36888701
I also tried other characters and although it is not clear for every image and character, there seems to be a general trend that v5.0 at least tends to mess up some character details compared to v4.0.
Awesome! The combination between realism and Pony-features is brilliant.
Somehow a lot of Lora's is causing it to get nuked.
Like for example: https://civitai.com/models/457418?modelVersionId=509163
Will result in a black image with red burned in spots.