如何使用构建器
从片段库添加内容，然后检查生成结果。一个稳定的提示词通常按主体、姿势或动作、服装、场景、镜头、光线、氛围、风格和质量词来组织。
提示词篮子保存在浏览器本地存储中，刷新后仍会保留。冲突检查只是建议，不是规则。
站立
放松的全身站姿，轮廓稳定清楚。
AI Image Prompt Generator
从可复用提示词库里选择姿势、场景、镜头和模型方向。
从片段库中选择内容，右侧会生成 Positive / Negative / Parameters。冲突会被标红提醒，但不会阻止复制。
姿势
姿势库
包含站立、行走、坐姿、跪姿、回头和动态站姿等核心身体姿势。
这个套组包含: 控制角色在画面中如何站、坐、移动或转身的身体位置提示词。
行走
自然向前行走，手臂和步伐容易读懂。
坐姿
端正坐姿，手部放置自然。
跪姿
单膝跪地，身体平衡稳定。
回头
三分之四身体扭转，优雅地回头看。
动态站姿
一脚向前、适合动作感画面的稳定站姿。
模型推荐入口
在复制到 ComfyUI 之前，用这些说明判断该选哪个 checkpoint。
checkpointmeinamix_v5Final
main anime sample library / pose cards / outfit examples / clean character prompts
Use Danbooru-style short tags plus plain English structure. Good first choice for standing/walking/sitting pose cards.
checkpointDarkSushiMix-2.25D
2.5D anime look / lighting comparison / stylized portraits / model comparison cards
Works well with anime character tags, lighting words, and camera framing. Keep realism tags moderate.
checkpointmajicmixRealistic_v7
realistic portrait comparison / camera tests / lighting tests / semi-photo examples
Use natural language prompts, explicit camera/lens/lighting, and fewer anime tags.
checkpointjapaneseStyleRealistic_v20
Japanese realistic style / soft portrait samples / realistic clothing materials / model comparison cards
Use natural language with Japanese street/interior styling notes. Avoid piling anime-only tags.
checkpointIllustrious-XL-v1.0
optional XL quality tests / high detail anime examples / later premium sample cards
Use XL-friendly prompts with clear subject, pose, style, and quality. Avoid using it for every first-pass sample.
checkpointQwen-Image-Edit-2511-Lightning-4steps
image editing / variation cleanup / turning approved samples into refined variants
Treat as an edit/refine tool rather than the main prompt library generator.lorasdpose_wholebody_fp16
pose control / whole-body reference guidance / standing/walking/sitting consistency
Use alongside OpenPose / pose reference workflows when you want the exact body pose to stay stable.