<h1 id="heading-1221">Edit(12/4/2024):</h1>Emm just don't use the LoRA w/t LayerDiffusion if you want better resultsAnd I can't really tell the difference between epoch 9 and epoch 10If you ask me I prefer epoch 9<h1 id="heading-1221">Edit(10/4/2024):</h1><h2 id="heading-269">Added ver. 2.4</h2>The prompt used in training has changed a bitYou may refer to the prompt used in samplesExt: LayerDiffusion is quite useful for testing, but seems like it would affect details a bit<h2 id="heading-1222">Adetailer</h2>can produce a better quality of face(eyes especially, the eyes look more similar to "spirit pupils" appear in Date A Live works) <h2 id="heading-1223">LoRAs recommend to use together</h2>You may also use with Add More Details LoRA by Lykon to create background with more details Here are some recommend prompts for background: scenery, day, daytime, street, background_sky So just describe the background you would like Origami to appear at as you like :)<h2 id="heading-1224">Generation Settings</h2><ul><li>Width: 512</li><li>Height: 768</li><li>Sampling Method: Eular a or DPM++ 2M Karras</li><li>Sampling steps: 20-40</li><li>Hires.fix settings: Upscaler: R-ESRGAN 4x+ Anime6B, Hires Steps: 20, Denoising Str 0.4 - 0.55, Upscale: 2</li><li>Clip Skip: 2</li></ul><h3 id="heading-1225">ADtailer Settings</h3><ul><li>Use face_yolov8s for face fix Threshold ,Denoising strength set to 0.55</li></ul><h3 id="heading-1226">Trigger Words: tobiichi origami angel</h3>Suggested prompts to be used with trigger words:1girl, bangs, bare shoulders, blue eyes, breasts, center opening cleavage, crown, detached sleeves, elbow gloves, hair between eyes, looking at viewer, medium breasts, portrait, short hair, simple background, sky, small breasts, smile, solo, white bow, white bowtie, white dress, white footwear, white gloves, white hair, white thighhighs, white veilTrained on animefull-latest.ckptRecommended Wright:0.8-1The training data includes other epochs that are not published(cause the other epochs might have problems in reproduce the details of training dataset, either missing or wrong) Difference between 000002 and 000004 are the training steps and epochsFeel free to try out 000004 as well(it has similar performance to 000002 according to my observations)

Date A Live | Tobiichi Origami

Files with SHA256: 87db4958277a1df9263f20e2ddf145bd2f6378ef749f9ccdc85a3c93c6daa9e4