About V1
60k high-quality male focused images were used in the new dataset, increasing its diversity. Unfortunately, many of the characters that were planned to be added were not learned well, but I intend to fix this in the next versions.
Recommended settings
CFG Scale: 5-7, Euler a
25-28 steps
Note: using lower cfg scale may cause body deformations and reduce details
Positive prompt :
masterpiece, best quality, absurdres, hi resYou may also use tag ai generated in positive or negative prompts
Negative prompt :
low quality, worst quality, lowres, username, sketch, censor, blurry, distorted, bad anatomy, signature, watermark, patreon logo, artist name
This is an experimental fine-tune of Illustrious V1.0, with a focus on yaoi and male characters. My goal was to create a model that could generate some of the newer, popular male characters from 2025.
The training dataset was quite small, around 7,000 images, focusing exclusively on some male characters. I also did a small u-net only rain with an additional 1,500 images to incorporate a bit of my original style. This is a small fine-tune and one of my first attempts at creating a model, please consider this an experimental version.
CFG Scale: 6+, Euler a
You may encounter anatomical problems, such as issues with fingers, eyes, and other details. To help mitigate some of these imperfections and any general blurriness, I highly recommend using the Hi-Res Fix feature during generation. I'm still learning, so the results might not be perfect, but I hope you enjoy experimenting with it!
Description
FAQ
Comments (2)
Your model is truly wonderful 🌟🔥🔥 – I love the art style, and it generates amazing male characters 😍 that are exactly to my taste. I wanted to give you this 801BUZZ to show my appreciation for all your hard work.⚡
When I use multiple LoRAs together, however, the anatomy tends to break down quite heavily, and adjusting it to match my preferred details takes a lot of effort . I’ve been experimenting with XYZ plot for about a week to study this. This issue is mostly because I use many LoRAs at once, but compared to other popular models, the anatomy breakdown happens more often here.
If that part could be improved, I think I could happily keep using this model forever ♾. Thank you for creating such a great work 🙏✨
A question for you: You said that a CFG lower than 5-7 can cause body deformations, but most of the time I generate with the Euler a DY CFG++ sampler. It's a dynamic sampler that performs some magic in the middle of the generation process to control itself, but one of the things about it is that using a CFG in that range can cause heavy burning and worse. Most of the time, I use it in the 1.2 to 2.5 range. Do you know whether maybe the body deformations won't apply so much? Another quirky thing about the dynamic samplers is that a lower CFG tends to give sharper results, just the opposite of normal. It's fine if the answer is "Try it and see", but I wanted to raise that point and at least make sure you're aware that some samplers need CFGs in a much lower range and so that might be a problem for them. Euler a isn't the only dynamic sampler, either. There are many of them.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.