콘티 (작화)

이 카테고리는

Composition tags control the overall artistic direction and rendering quality of the generated image. They sit at the very beginning of effective prompts because the model uses them to pick which distribution within its training set to sample from. The category covers three related axes: quality markers (masterpiece, best_quality, absurdres, highres) that signal top-tier output; art style tags (anime, realistic, pixel_art, watercolor) that choose the visual medium; and camera / art type modifiers that refine either. If you remove every other category from your prompt but keep composition, the image will still have a clear aesthetic identity — that is how foundational this axis is. Most prompts start with two or three composition tags before naming the subject, though our own fixed-seed comparison on an Illustrious checkpoint found almost no visible difference between worst quality and a full masterpiece stack, so treat quality markers as a cheap convention rather than the biggest lever in the prompt. The tags here are model-agnostic in name but are tuned primarily for Stable Diffusion family models (SDXL, SD 1.5, NovelAI, Pony fine-tunes). Flux and Midjourney still respond to many of them, though they weigh them less heavily than Danbooru-trained models do.

전체 38개 태그 — 하위 분류별

관련 카테고리

These categories pair naturally with composition tags — adding one tag from each builds a more cohesive prompt.