Clean-line characters surrounded by home-built machines. Think flat cel colour with one shadow tone, thin dark outlines, pink-blushed cheeks, and plain backdrops in mustard, mint, coral or dusty pink. The signature move is a ring of retro gadgets floating around the figure, joined by coiled black cables: speakers, gauges, riveted boxes, plugs, glowing canisters and katakana labels.Trigger word: <code>k0r1ki</code>How to use itStart every prompt with <code>k0r1ki, </code>. After that, describe the picture in plain sentences, the way you'd caption it:<ol><li>The character's face: hair, eye colour, brows, nose, mouth, expression</li><li>What she's doing, and her clothes</li><li>The props her job or hobby would give her</li><li>A plain background colour ("Plain mint-green background.")</li></ol>Leave out style words such as "anime", "flat colours" or "illustration". The trigger carries the look, and those words pull the base model back toward its own anime style.The floating machinery only shows up when you ask for it. Name the gadgets you want ("floating around her are a teal speaker, a pressure gauge and a coiled cable") and leave them out when you want a figure alone.SettingsKrea 2 Turbo: weight 1.0, 8 steps, guidance 1. Raise the weight to 1.2 if the faces come out glossy. Portrait 896 × 1344, landscape 1344 × 896.Tips<ul><li>Give every face a quirk in the brows, nose or mouth: thick straight brows, a long pointed chin, a lopsided half-smile. Otherwise every face drifts toward the same blushing bob.</li><li>If a figure stands next to a big object, say which way she faces and where her feet are, and add a ground shadow. Without that, Krea sometimes turns one foot backwards.</li><li>Scenes with no character in them get a weaker effect. Only one training image had no person in it.</li></ul>Training27 hand-picked images, each captioned by hand with content only: faces, clothes, props and background colour, with no style words. Trained on Krea 2 Base with Civitai's trainer: 2,160 steps over 10 epochs (about 80 passes per image), dim 32, alpha 32, UNet LR 1e-4, constant scheduler, AdamW8bit, flip augmentation off.

Files with SHA256: 08e2447de4fcd424adbace7625e3f47cf4bd598a0bb4c12518e2b6150f86a848