I trained this LoRa with training images from this <a rel="ugc" href="https://civarchive.com/models/534506?modelVersionId=1518834">Pony Lora</a>.Activation Prompt: N1j1Negative Prompt: lowres, worst quality, bad quality, bad anatomy, sketch, jpeg artifacts, (signature), (watermark), (text), (artist name), logo, old, oldest, chibi, loli, breasts ptosis, (censored), bar censored, mosaic censored, bb \(baalbuddy\), (multiple views)*This ^^^ negative prompt works best.To get the same results like the sample images follow this prompting formula: Your Prompt + BREAK Prompt + LoRaBreak Prompt: BREAK character_concept, cowboy_shot, (dutch_angle:1.25), masterpiece, best quality, amazing quality, very aesthetic, absurdres, modern, recent, newestPrompt Example: 1girl, solo, virgin ribbed sweater, destroyer sweater, dark violet green sweater, navel, thighhighs, smile, looking at viewer, library, thick thighs, wide hips, large breasts, curvy, BREAK character_concept, cowboy_shot, (dutch_angle:1.25), masterpiece, best quality, amazing quality, very aesthetic, absurdres, modern, recent, newest, &lt;lora:NijiSemi_IL_Style_epoch_8:1&gt; N1j1*This prompt formula works best if you are using reForge or StableDiffusion locally. Based on my tests it does not work as good if you generate them here on the site.Weight: 1Sampler: Euler a - AutomaticRecommended Base Models: <a rel="ugc" href="https://civarchive.com/models/1132549?modelVersionId=1294597">RoseMIX_XL</a>Make sure to post your favorite generations on the Gallery by simply clicking +Add Post button.

Files with SHA256: 38478dc63635193cf4ccc5b322cab3e1328ebf61a69e919ac864cd5164477b0b