女游戏 influencer 2×2：可复用 influencer_frame 模板

提示词

2x2 grid, 16:9, do this for 4 female gaming influencers with poses that steal your heart. // FN: influencer_frame(SUBJECT, SETTING, TIME, RIG, PLATFORM, INTENT)  // Edit INPUT only. Everything below derives.  INPUT   SUBJECT  = " $ subject"    SETTING  = "Ai infers"    TIME     = "Ai infers"   RIG      = SELF            // SELF | FRIEND | TRIPOD | MIRROR | DRONE   PLATFORM = IG_FEED         // IG_FEED | IG_STORY | TIKTOK | PINTEREST   INTENT   = "candid morning"   INFER :: RIG → geometry, gaze, and what the fiction is   SELF    → 23mm front, 45–60cm, face widened ~8%, lens ABOVE eye level,             chin foreshortened, gaze INTO lens, one arm's angle implied by crop   FRIEND  → 26mm rear, 1.5–3m, correct proportion, eye level, gaze into lens,             full body possible, environment legible   TRIPOD  → 26mm, 2–4m, gaze deliberately AWAY (the candid fiction), body             squared to a mark on the ground, static pose held for a burst   MIRROR  → phone IN FRAME, screen glow on face/hands, gaze at reflection NOT             lens, room reversed, fingerprints on the glass   DRONE   → 24mm from above, subject small, the LOCATION is the subject  INFER :: SETTING + TIME → light class → computational behavior   interior_backlit  → HDR stack: window keeps detail, never clips; edge halo   golden_hour       → warm, long shadows, veiling flare, lifted blacks   overcast          → flat, cool, low contrast, no shadow direction   night_artificial  → night mode: multi-second stack ⇒ anything that moved                       ghosts; shadows unnaturally lit; grain scrubbed to plastic   harsh_noon        → HDR fights it: raccoon shadows lifted, skin looks flat  INFER :: PLATFORM → container   IG_FEED  = 4:5,  1080px ceiling, mild re-encode   IG_STORY = 9:16, safe zones top/bottom kept empty   TIKTOK   = 9:16, heavier compression, saturation pushed   PINTEREST= 2:3,  text overlay space reserved  DERIVE :: always, regardless of input   D1 synthetic depth ⇒ mask errors mandatory: one hair-gap filled with blur,      a ~2px halo at the shoulder, blur bucketed by depth not continuous   D2 beauty pass ⇒ cheeks poreless WHILE lashes/brows stay razor sharp   D3 grade ⇒ blacks lifted to ~12/255 with cyan bias, nothing pure black   D4 INTENT ⇒ props staged: labels rotated to camera, food untouched,      "unnoticed" objects squared to the table edge  FORBIDDEN   ✗ true optical bokeh falloff        ✗ clipped white window   ✗ uniform skin sharpness            ✗ pure black   ✗ DSLR micro-detail at 1080         ✗ genuine unawareness of the camera  RENDER  // resolve every {slot}, emit prose only   {PLATFORM.aspect} phone photo. {SUBJECT} at {SETTING}, {TIME}.   {RIG.geometry_prose}. {light_class.prose}. Background blurred the artificial   way phones do it, with {D1}. {D2}. {D3}. {D4}. Slightly soft at 1080px.  AUDIT   window_luma_max < 250 · cheek_highfreq << lash_highfreq   ≥1 segmentation halo · black_point > 8/255 · gaze matches RIG

中文译文

2x2 网格，16:9，为 4 位女游戏 influencer 拍这组图，姿势要能偷走你的心。 // FN: influencer_frame(SUBJECT, SETTING, TIME, RIG, PLATFORM, INTENT)  // 只编辑 INPUT。以下全部为推导结果。  INPUT   SUBJECT  = " $ subject"    SETTING  = "Ai infers"    TIME     = "Ai infers"   RIG      = SELF            // SELF | FRIEND | TRIPOD | MIRROR | DRONE   PLATFORM = IG_FEED         // IG_FEED | IG_STORY | TIKTOK | PINTEREST   INTENT   = "candid morning"   INFER :: RIG → 几何关系、视线，以及这个“虚构”究竟是什么   SELF    → 23mm 前置，45–60cm，脸部拉宽约 8%，镜头高于眼平线，             下巴透视压缩，视线看向镜头内，手臂角度由裁切暗示   FRIEND  → 26mm 后置，1.5–3m，比例正确，眼平高度，视线看向镜头，             可拍全身，环境清晰可辨   TRIPOD  → 26mm，2–4m，视线刻意避开镜头（偷拍感的虚构），身体             正对地面上的标记点，静态姿势连拍   MIRROR  → 手机在画面内，屏幕光映在脸/手上，视线看镜面倒影而非             镜头，房间左右反转，玻璃上有指纹   DRONE   → 24mm 俯拍，主体很小，地点才是主体  INFER :: SETTING + TIME → 光线类别 → 计算摄影行为   interior_backlit  → HDR 合成：窗户保留细节，绝不过曝；边缘光晕   golden_hour       → 暖调，长阴影，雾状炫光，黑位抬升   overcast          → 平，冷调，低对比，无阴影方向   night_artificial  → 夜景模式：多秒合成 ⇒ 任何移动过的东西                       都会拖影；阴影被不自然地照亮；噪点被抹成塑料感   harsh_noon        → HDR 与它对抗：浣熊眼阴影被抬亮，皮肤显得扁平  INFER :: PLATFORM → 画幅容器   IG_FEED  = 4:5,  1080px 上限，轻微二次压缩   IG_STORY = 9:16，顶部/底部安全区留空   TIKTOK   = 9:16，压缩更重，饱和度推高   PINTEREST= 2:3,  预留文字叠加空间  DERIVE :: 无论输入为何，始终执行   D1 合成景深 ⇒ 遮罩错误是必须的：一处发丝缝隙被模糊填满，      肩部约 2px 的光晕，模糊按景深分层而非连续过渡   D2 美颜处理 ⇒ 脸颊无毛孔，而睫毛/眉毛保持刀锋般锐利   D3 调色 ⇒ 黑位抬至约 12/255 并偏青，没有任何纯黑   D4 INTENT ⇒ 道具摆位：标签转向镜头，食物未被动过，      “未被注意”的物体与桌边对齐  FORBIDDEN   ✗ 真实的光学焦外衰减        ✗ 窗户高光过曝死白   ✗ 皮肤锐度均匀            ✗ 纯黑   ✗ 1080 下的单反级微细节         ✗ 对镜头真正的毫无察觉  RENDER  // 解析每一个 {slot}，只输出文字描述   {PLATFORM.aspect} 手机照片。{SUBJECT} 在 {SETTING}，{TIME}。   {RIG.geometry_prose}。{light_class.prose}。背景以手机那种人造的方式  虚化，带有 {D1}。{D2}。{D3}。{D4}。在 1080px 下略微偏软。  AUDIT   window_luma_max < 250 · cheek_highfreq << lash_highfreq   ≥1 segmentation halo · black_point > 8/255 · gaze matches RIG

溯源

原作者

Gdgtify

发布日期

2026-09-17

原始出处

查看原帖 ↗

收录来源

xianyu110/awesome-gpt-image2.5 · 2100581539008626771 · MIT