``` Kivotos 3.0 V-Pred is an attempt to convert the original SDXL model into a v-prediction type model, with the goal of enabling true dark and true light generation. ==================== USAGE GUIDE ==================== Model Type: Hybrid (supports both tags and natural language prompts) Prompt Structure: - Tags: 1girl, solo, haruna, hews, looking at viewer, night, dark, light smile - Natural: A girl named Haruna from Blue Archive, in the style of Hews, looking at the viewer with a light smile in the middle of the night - Hybrid: 1girl, haruna, blue archive, in the style of Hews, she is looking at the viewer and showing a light smile, dark, night ==================== SPECIAL TAGS ==================== Same as Moescape Diffusion 1.0 - Quality tags: masterpiece, best quality, low quality, worst quality - Score tags: high score, great score, good score, average score, bad score, low score - Temporal tags: year 2005, year 2015, year 2020, ... year n - Rating tags: safe, sensitive, nsfw, explicit ==================== RECOMMENDED SETTINGS ==================== Quality prompts: masterpiece, high score, great score, absurdres, sensitive Negative prompt: Use the recommended negative prompt Sampler steps: 25–30 (28 recommended) CFG scale: 4–7 (5 recommended) Preferred samplers: Euler or Euler A Training resolution: 1024 × 1024 Supported resolutions: - 1024 × 1024 - 1152 × 896, 896 × 1152 - 1216 × 832, 832 × 1216 - 1344 × 768, 768 × 1344 - 1536 × 640, 640 × 1536 ==================== STYLE SUGGESTIONS ==================== faux traditional media, watercolor, oil painting, impressionism, impasto, abstract, chiaroscuro, fine art parody, art nouveau, etc. ```