<h2 id="hello-everyone!">Hello everyone!</h2>I'd like to introduce my new LoRa designed for decensoring images. It is primarily trained for Anime styles, but it can also be used on 3D and photorealistic images (though for realism, it may result in a somewhat "plastic" skin texture).This LoRa is capable of reconstructing anatomical details such as veins and bodily fluids on penises, as well as vaginas and anuses. While the results aren't always flawless, it successfully restores specular highlights if they are present in the overall art style.Base Model: Qwen Image Edit 2511Capabilities:<ul><li>Censorship types removed: Mosaic censor and bar censors over genitals.</li><li>Note: The softer/lighter the mosaic, the better the final result will be.</li></ul>This is version v0.5, meaning it's still a work in progress and not perfectly accurate yet, but I specifically designed it to be highly flexible across various art styles.Testing &amp; Compatibility: All my tests were conducted using the original <a rel="ugc" href="https://huggingface.co/Comfy-Org/Qwen-Image-Edit_ComfyUI/tree/main/split_files/diffusion_models">FP8 diffusion model for ComfyUI</a>, which yields the best results. However, it is also fully compatible with <a rel="ugc" href="https://huggingface.co/Phr00t/Qwen-Image-Edit-Rapid-AIO">Phr00t's AIO NSFW</a> model and works well alongside <a rel="ugc" href="https://huggingface.co/lightx2v/Qwen-Image-Edit-2511-Lightning/tree/main">Lightning LoRas</a>.Training Details:<ul><li>Dataset: 500 image pairs</li><li>Steps: 20,000</li><li>LoRa Rank: 64</li><li>Learning Rate: 0.00008</li></ul>Trigger word: <code>remove censor</code>I will prepare and post usage examples below shortly.A Note on Future Updates: I would absolutely love to continue improving this LoRa, but unfortunately, the training process is very expensive for me. "Edit" models take significantly more time to train because they need to process multiple images simultaneously. Because of these costs, this might be the final version of this particular LoRa.That being said, I still have a lot of ideas for other LoRas I want to create!

Files with SHA256: dcf4dc4a387edc610216e9a63e2c8d05ba92c6551ef9b004f330536f3db3d073