PROMPT RADAR / IMAGE插画三维
3D character artists should be more worried about Kimi K3 than Midjourne
3D character artists should be more worried about Kimi K3 than Midjourney. Image generators can fake one pretty frame. The real threat is AI taking a dead base mesh and handling the painful middle: hair, clothes, materia
@irinatoxi 1.3k 热度 2.8 万 浏览
编辑先看
它为什么值得进入灵感库
先别急着照抄整段 Prompt：真正有用的是主体、空间和镜头之间的约束关系。 轮廓语言、材质规则和世界观要互相支持，避免把互不相干的风格词堆在一起。
01 / 原始方法
完整视觉 Prompt
3D character artists should be more worried about Kimi K3 than Midjourney. Image generators can fake one pretty frame. The real threat is AI taking a dead base mesh and handling the painful middle: hair, clothes, materials, lighting, posing, camera work, and fifty tiny fixes nobody posts online. That is exactly where workflows around Kimi K3 are heading. Not “generate an anime girl,” but keep refining the same editable 3D scene until it stops looking like a mannequin and starts looking marketable. When that works reliably, clients will stop paying for hours of technical suffering. They will pay for taste. And a lot of portfolios are not going to survive that distinction.
02 / 复刻路径
不要一次改完所有变量
用 Midjourney 复刻时，先固定主体和构图，再逐次替换光线、色调或材质中的一个变量，这样更容易判断哪项描述真正起作用。
- 主体：3D character artists should be more worried about Kimi K3 than Midjourney. Image generators can fake one pretty frame. The real threat is AI taking a dead base mesh and handling the painful middle: hair, clothes, materia
- 视觉方向：实时 AI 视觉案例；查看原帖媒体与作者说明
- 镜头与构图：查看原帖画面；查看原帖媒体与工作流
- 交付约束：auto，保留原作者方法与归属；复刻时避免水印、低清、畸变、主体漂移和无关文字。
负向约束
保留原作者方法与归属；复刻时避免水印、低清、畸变、主体漂移和无关文字。
模型建议
Midjourney / GPT Image 2 / Grok Imagine / Flux。先用原始 Prompt 建立基准结果，再根据模型对自然语言、镜头运动或文字排版的能力做局部改写。
来源与编辑原则