<h3 id="heading-15"><strong>Put in your negative prompt:</strong></h3><h3 id="heading-16">&lt;lora:EasyFix:1&gt; (overfit style:1.0)</h3><p></p><p><strong>EasyFix </strong>is a negative LoRA trained on AI generated images from CivitAI that show <strong>extreme </strong>overfitting. This LoRA improves generated image quality without any major stylistic changes for any SDXL model.</p><p></p><p><strong>Suggested Strength:</strong><em> 1 to 16</em></p><p><strong><em>Important</em></strong><em>: adjust the strength of (overfit style:1.0) more than the strength of the LoRA</em></p><p></p><p></p><h3 id="heading-3998">What is overfitting?</h3><p>According to IBM (source: <a target="_blank" rel="ugc" href="https://www.ibm.com/topics/overfitting">https://www.ibm.com/topics/overfitting</a>), "<em>Overfitting is a concept in data science, which occurs when a statistical model fits exactly against its training data. When this happens, the algorithm unfortunately cannot perform accurately against unseen data, defeating its purpose. Generalization of a model to new data is ultimately what allows us to use machine learning algorithms every day to make predictions and classify data.</em>"</p><p></p><h3 id="heading-3999">What does overfitting look like in Stable Diffusion?</h3><p>Repeating patterns, nonsensical details, small dots, strange artifacts, facial blemishes... etcetera. If you look at an AI generated image and see some strange details that don't make sense, the chances are high that the image shows signs of overfitting.</p><p></p><h3 id="heading-1325">Overfitting as an Aesthetic</h3><p>Aesthetic appeal is subjective, so if you enjoy the incredibly detailed images Stable Diffusion is good at making, don't view this model as an attack on your personal taste! Please do an A/B comparison using EasyFix and judge for yourself which is better.</p><p></p><h3 id="heading-1325">Avoiding Overfitting</h3><p><strong>In Image Generation:</strong></p><p>Simpler prompts, with minimal negative prompting. If needed, lower CFG scale.</p><p></p><p><strong>In Training:</strong></p><p>I will eventually be writing a short guide for advanced users on how I train my LoRA, so watch this space. AI_Characters has an excellent guide on training methodology (see: "Evaluating your model") <a target="_blank" rel="ugc" href="https://civarchive.com/articles/1771">https://civarchive.com/articles/1771</a></p><p></p><h3 id="heading-733">Methodology</h3><p>Collect tagged AI-generated images from CivitAI that show overfitting (using <a target="_blank" rel="ugc" href="https://github.com/hassan-sd/civitai-image-scraper">https://github.com/hassan-sd/civitai-image-scraper</a>). Images should be greater than minimum size of 768x768.</p><p>Remove all brackets and attention strength from captions. (e.g. "((cute dog:4))" becomes "cute dog"). Remove all tags relating to quality (e.g. "masterpiece, best quality, insanely detailed"). Add a trigger word at the start of each word like "overfit style".</p><p>Train on the images and generate A/B comparison images while training to save time later.</p>

EasyFix [Negative LoRA] SDXL

<p>Trained for 38 epochs. Trained with input noise perturbation 0.1 <a target="_blank" rel="ugc" href="https://github.com/kohya-ss/sd-scripts/pull/798">https://github.com/kohya-ss/sd-scripts/pull/798</a>. Prodigy with CosineAnnealingLR with T_Max set to max epochs (50). LoRA-FA using kohya-ss. Network rank/alpha: 8/4, Conv rank/alpha: 1/1</p>

Files with SHA256: a85bdf62bc95c69c52dca7d40559dfce928d6dddd339ca51735021b34e4d573f