Testing loras on civitai. Models might be hit or miss since I can't test before publishing.I created this model to have a single LoRA that has a style similar to these images: <a target="_blank" rel="ugc" href="https://civarchive.com/images/20883269">image 1</a>, <a target="_blank" rel="ugc" href="https://civarchive.com/images/20899834">image 2</a> (NSFW), <a target="_blank" rel="ugc" href="https://civarchive.com/images/20878432">image 3</a> (NSFW).<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/76cbd13e-17db-4cc5-a426-0f959d31ed66/width=525/76cbd13e-17db-4cc5-a426-0f959d31ed66.jpeg" />Model was training using generated images from a combination of the following:<ul><li>SD model - <a target="_blank" rel="ugc" href="https://civarchive.com/models/257749/pony-diffusion-v6-xl?modelVersionId=290640">Pony XL</a></li><li>LoRa (style) - <a target="_blank" rel="ugc" href="https://civarchive.com/models/319503/suoiresnu-lora-or-sdxl-pony?modelVersionId=358270">Suoiresnu</a></li><li>LoRa (style) - <a target="_blank" rel="ugc" href="https://civarchive.com/models/341353/expressiveh-hentai-lora-style?modelVersionId=382152">ExpressiveH</a></li><li>LoRa (concept) - <a target="_blank" rel="ugc" href="https://civarchive.com/models/359677/blackedwear-or-concept-lora-or-experimental-or-sdxl-pony?modelVersionId=402001">Blackedwear</a></li><li>LoRa (style) - <a target="_blank" rel="ugc" href="https://civarchive.com/models/362449/bartolomeobari-style-pony-xl?modelVersionId=409077">Bartolomeobari</a></li><li>LoRa (style) - <a target="_blank" rel="ugc" href="https://civarchive.com/models/340248/sinfully-stylish-dramatic-lighting?modelVersionId=449162">Sinfully Stylish</a></li><li>LoRa (style) - <a target="_blank" rel="ugc" href="https://civarchive.com/models/430143/fishine-hirame-ireading62-style-lora?modelVersionId=489794">fishine</a></li><li>LoRa (slider) - <a target="_blank" rel="ugc" href="https://civarchive.com/models/402462/detail-slider-lora-or-ponyxl-sdxl?modelVersionId=600608">Detail Slider</a></li></ul>Changelog:<ul><li>v1 - the model is workable, but not really what I wanted (see training data)</li><li>v2 - added more images and tags, switched to Adafactor<ul><li>simple images might need <code>halo</code> in the negative</li></ul></li><li>v3 - added a lot more variation to the images</li><li>v3.1 - same as v3, but trained with Prodigy and less repeats/steps - this one is my favourite</li><li>v4 (in progress) - will try to reduce overfitting</li></ul>

stylish LoRA (PonyXL)

Training: defaults (Prodigy), 412 auto tagged images (20 tags, 0.4 threshold)

Files with SHA256: 3f607caed9b64fac529bf2a51fd7aaf3e7d8cd04acba398a0f3949a71fdce6cf