MiniMax H3-Image 案例与 Prompt

一张 4K 暗黑奇幻动作 RPG 截图，定格首领战开始的瞬间，包含有名字的首领、血条和完整 HUD，渲染效果宛如真实引擎画面。

A hand-drawn, early-2000s style character customization web page built around an uploaded character design: species and item pickers, color dropdowns and a bio box. Needs your own character image as input.

A fake next-gen open-world game screenshot: flooded coastal streets, floating cars, a police helicopter and lightning over the bay, with in-game HUD. One structured prompt covers scene, camera, weather and UI.

A hyper-real smartphone selfie of a young woman on a high-rise hotel balcony with the Hong Kong skyline at night. The author ran the same prompt on four models; this is the MiniMax H3-Image output (first image of the post).

一张超现实梦核摄影作品：一名身穿办公服装的男子坐在无边无际的白色花田中央，在办公桌前打字。出自 MiniMax 官方 H3-Image 帖子。

一组 3D 游戏资产包展示：深色摄影棚背景前，摆放着红瓦顶中世纪小木屋、木桶、木箱和石灯，并配有线框示意面板。提示词着重要求 PBR 纹理和摄影棚布光。

A film-noir detective evidence board that actually solves a six-suspect true/false logic puzzle: portraits, statements marked true or false, the culprit highlighted and a verification table. From MiniMax's official H3-Image thread on turning information into diagrams.

将现代宇航员维修空间站的场景，以矿物颜料绘成唐代敦煌土壁壁画，采用飞天式构图。出自 MiniMax 官方 H3-Image 帖子。

A photoreal overhead shot of a printed services agreement on a desk, with nearly 5,000 characters of legible contract text generated from the prompt alone. A stress test for long-form text rendering.

A mixed-media poster: photoreal glass, water and ice with tiny hand-drawn pencil characters living inside, plus Chinese handwritten captions. The prompt is a fill-in template with variables; shown is Bubble Diving.

一幅 9:16 动漫插画，采用随性的墨线和低饱和灰紫色调：白发剑士身穿宽大连帽衫，手握武士刀，以 extreme low angle 呈现。帖子分享了四条提示词，这是第一条。

A cinematic 1940s train heist movie poster that places eight referenced characters, four props, a setting, a lighting and a grade reference, plus a title ticket, in one frame. Needs your own 16 reference images.

A dynamic 2D animation-style illustration of a young psychic boy with glowing white eyes and crackling energy in his hands, prompt-only with no reference images.

一张 9:16 时尚杂志封面，使用 direct flash，模特身后有一个超大字母。提示词设定了四种色彩方向（PULSE、THROUGH、HALO、CHIC），要求每次生成一张封面；展示的是 PULSE。

一张具有电影感的 slow-shutter panning 摄影作品，描绘林荫道路上的骑行者，融合复古体育纪实风格与 35mm 胶片质感。作为创作者作品分享于 MiniMax 官方 H3-Image 帖子中。

A dark-fantasy AAA environment concept: a ruined Gothic cathedral rising from a flooded basin, a colossal eroded saint statue and a lone armored wanderer for scale. The post shares four environment prompts; this is the first one, matching the first image.

MiniMax H3 Image：用截图作参考生成 16:9 动画角色设定表的完整提示词（正/侧/背全身与面部、眼睛特写）；提示词在作者自回复，附对照示例图。

MiniMax H3 Image：超写实儿童手写信给上帝俯拍摄影提示词（含完整信件正文、笔迹与涂鸦要求）；完整提示词在作者自回复，附示例图。

🚨 中国又推出了一个惊人的模型.. MiniMax H3 Image 已开放抢先体验，塑料皮肤质感消失了，输入 1 张照片.. 输出完整 4K 角色设定图.. 毛孔…