This model is designed to generate beautiful female anime characters in a highly consistent anime style from 2015–2025, closely resembling the clean and elegant aesthetic of Kaguya-sama: Love is War (A1 studio style).<h3 id="what-you-can-expect:">What you can expect:</h3><ul><li>Detailed, and highly consistent anime girls</li><li>Perfectly matching anime-style backgrounds (no more style clashes)</li><li>Excellent control over lighting — from soft and dreamy to dramatic and cinematic</li></ul>No special trigger word required! Just include "anime" and "anime girl" in your prompt and it works beautifully.The entire dataset was captioned in natural, fluent English with zero tags — making this LoRA especially beginner-friendly.<h3 id="recommended-usage:">Recommended usage:</h3><ul><li>NSFW: This LoRA was trained only on SFW/clean images. To generate NSFW content, combine it with a good NSFW LoRA. I personally tested it with <a target="_blank" rel="ugc" href="https://civarchive.com/user/Ashen3">SNOFS</a>.</li><li>For best results, add at the end of your prompt: Detailed anatomy, perfect anatomy for "the important body parts".</li></ul><h3 id="img2img:">Img2Img:</h3>This LoRA excels at style transfer. You can transform any image — even realistic photos — into this anime style.Simple prompt formula: "In the style of anime"Note on character consistency: Skin tone and eye color may shift slightly during Img2Img. To preserve them better:<ul><li>Directly describe in the prompt (e.g. fair skin, blue eyes)</li><li>Or add: Preserve subject's facial features, skin tone, expression, figure and pose. Preserve overall composition of this photo.</li></ul><h3 id="technical-details:">Technical details:</h3>The LoRA was created through sequential merging of three LoRAs:<ul><li>First two base LoRAs: <a target="_blank" rel="ugc" href="https://civarchive.com/models/2486624/hayasaka-ai-flux2-klein-9b-character-lora">Link 1</a> + <a target="_blank" rel="ugc" href="https://civarchive.com/models/2491642/shinomiya-kaguya-flux2-klein-9b-character-lora">Link 2</a></li><li>The third was an identity + anatomy fix, trained on approximately 88 images generated from the merged first two</li><li>Total dataset size: ~230 images</li></ul>The model performs excellently with female characters. It can generate males "in some anime style," but I do not recommend using it for male characters until a proper update or dedicated male version is released.

Files with SHA256: f82bdf186138681632c35a6433e8b020871c3d81c50d1ae765a00cb449a3350e