It takes substantial time and efforts to bake models. If you appreciate my models, I would be grateful if you could support me on <a target="_blank" rel="ugc" href="https://ko-fi.com/koolchh">Ko-fi</a> ☕ .<h2 id="heading-2">Husbando LoCon is a LyCORIS for creating bara/yaoi hentai and sexually suggestive males.</h2>🚀 Please also check out <a target="_blank" rel="ugc" href="https://civarchive.com/models/208697/selai-mix">Selai Mix</a>, my recent checkpoint merge model release, if you love the aesthetics of Husbando LoCon!<h3 id="heading-1152">Features</h3><ul><li>Easy to create good-looking masculine characters using tag-based prompts.</li><li>Good male genitalia.</li><li>Good solo / solo focus homoerotic content.</li><li>Bara/yaoi hentai depicting 2 or more people are highly erroneous. You'll need luck and careful prompting to get things right.</li></ul><h3 id="heading-1153">Recommended Settings (v4)</h3><ul><li>CLIP skip: 2</li><li>You can start by trying a weight of 0.5 with CFG scale 7.</li><li>For hentai art, I recommend LoRA weight of 1.0 with CFG scale 9.</li><li>Since <a target="_blank" rel="ugc" href="https://civarchive.com/models/208697/selai-mix">Selai Mix</a> already contains Husbando LoCon v4.0, a lighter weight such as 0.3 should already give you the desired effect. You should not set the weight to over 0.7 when using on Selai Mix.</li></ul><h3 id="heading-1154">How to use</h3><ul><li>Select any checkpoint that works with tag-based prompts. Recommended models: <a target="_blank" rel="ugc" href="https://civarchive.com/models/208697/selai-mix">Selai Mix</a> and <a target="_blank" rel="ugc" href="https://civarchive.com/models/23900/anylora-checkpoint">AnyLoRA</a>.</li><li>Select a VAE if the checkpoint model you're using doesn't have a baked-in VAE, or else you'll find your generated images have dull colors. Recommended VAEs: <a target="_blank" rel="ugc" href="https://huggingface.co/NoCrypt/blessed_vae/tree/main">blessed2</a>, <a target="_blank" rel="ugc" href="https://huggingface.co/stabilityai/sd-vae-ft-mse-original/tree/main">vae-ft-mse-840000-ema-pruned</a> or <a target="_blank" rel="ugc" href="https://huggingface.co/hakurei/waifu-diffusion-v1-4/tree/main/vae">kl-f8-anime2</a>.</li><li>Make sure to follow the steps in <a target="_blank" rel="ugc" href="https://github.com/civitai/civitai/wiki/How-to-use-models#lora">How to use a LoRA</a>.</li><li>Generate at 512 base resolution (512x768, 512x512 or 768x512) and use hires fix to improve the details. Recommended upscaler: R-ESRGAN 4x+ Anime6B; denoising strength: 0.4.</li></ul>

Husbando LoCon

<ul><li>Fixed the overfitting problem in previous versions.</li><li>Even larger training dataset (6,617 images)</li><li>Auto tagged with wd-v1-4-moat-tagger-v2 with zero pruning.</li><li>Used flip augmentation and color augmentation.</li><li>Used multi resolution noise during training.</li><li>Obtained from LoCon extraction after fine-tuning the entire model.</li><li>Downscaled the LoCon dimension to 128.</li></ul>

Files with SHA256: cca9887784c794631009df7cf10000e5d17ffabcc9e9aacec28936e329349911