這套題目是為 HiDream-O1 官方宣稱驗證那篇寫的,跟我之前幾篇評測的題庫完全不重疊。它不是泛用題庫,每一章都針對一條官方宣稱設計,所以拿去測別的模型時,看的是「這個模型在這種需求下行不行」,而不是總分。
這套題目的設計邏輯
- A 章 8 題:每題把一種細小質感放在畫面中央(皮膚毛孔、織紋、霜晶、拉絲金屬、貓毛、繡線、珊瑚、麵包氣孔),專測有沒有 VAE 對細節的影響。裡面故意放了手錶這種精密物件,看模型會不會補出多餘的東西
- B 章 13 題:繁體中文 8 題,把「廟」「鹽」「麵」「龍」「讓」「龜」「鬱」「柵」「籲」「疊」「糰」這些筆畫複雜的字埋進招牌、看板、對聯這類實用場景;英文長文 3 題測整段逐字正確;混排 2 題測中英日並列與多區域版面
- C 章 9 題:編輯與合成的參考圖全部先用另一個模型生成,所以每一張都可以公開;編輯指令都加了「其他保持不變」,模型做不做得到這句話是重點
- D 章:兩題四格分鏡,格內有編號與標題要求;15 種鏡位用同一個主體、同一個 seed,只換鏡位描述
- E 章 8 題:故意寫得很短,而且要先知道某件事才畫得出來(詩句內容、牛頓是誰、汝窯長什麼樣);這章同時公開官方 Prompt Agent 改寫後的完整版本,可以直接比較兩種寫法
- F 章:三種官方訓練過的長寬比用同一個九份題目
使用建議
- 要求渲染的文字一律放在英文引號內,並描述字型、顏色、位置
- 每組同 seed 跑不同分支才能比較;文字題出圖後逐字核對,錯字常常長得很像
- E 章的短句直接餵模型通常畫不對,這是設計上的預期;想要正確結果,把 Agent 改寫版拿去用
A. 質感特寫
A01. 逆光雀斑特寫
Extreme close-up portrait of a young woman with dense freckles and fine vellus hair on her cheek, lit by low golden backlight from a window behind her, individual eyelashes and skin pores visible, loose strands of auburn hair catching the light, 100mm macro lens, shallow depth of field, photorealistic.
中文說明:年輕女性臉頰特寫,雀斑密集、絨毛可見,窗邊低角度金色逆光,睫毛與毛孔清晰,栗色髮絲被光勾邊,100mm 微距、淺景深、寫實。測皮膚這類天然質感的細節。
A02. Harris Tweed 花呢
Product photo of a folded Harris Tweed jacket sleeve on a walnut table, the herringbone weave with flecks of orange, moss green and heather purple yarn clearly resolved, a horn button with cross-stitched thread, soft window light from the left, high-resolution catalogue photography.
中文說明:摺好的 Harris Tweed 外套袖子放在胡桃木桌上,人字紋織法與橘、苔綠、石楠紫的雜色紗線要清楚,牛角鈕扣有十字縫線,左側窗光,型錄攝影。測織紋與雜色纖維。
A03. 霜楓葉
Fallen red and amber maple leaves on dark wet slate, edges rimmed with morning frost crystals, leaf veins and tiny ice needles sharply visible, overhead top-down shot, cold blue ambient light with one warm ray of sunrise, photorealistic macro.
中文說明:紅與琥珀色楓葉落在濕黑板岩上,葉緣結一圈晨霜,葉脈與冰針清晰,正上方俯拍,冷藍環境光加一道暖色晨光,寫實微距。
A04. 鈦金屬潛水錶
A brushed titanium dive watch on a slate coaster, radial brushing on the bezel, engraved minute markers, a scratched sapphire crystal with faint reflections, leather strap with visible grain and stitching, studio product photography with a single softbox, ultra sharp.
中文說明:拉絲鈦金屬潛水錶放在板岩杯墊上,錶圈放射狀拉絲、刻字分鐘刻度、藍寶石鏡面有刮痕與淡反光,皮革錶帶皮紋與縫線可見,單柔光箱棚拍。測精密物件會不會多長東西。
A05. 緬因貓毛髮
Close-up of a silver tabby Maine Coon cat sitting on a windowsill, long guard hairs and whiskers individually resolved, ear tufts backlit by afternoon sun, green eyes with visible iris texture, wooden window frame with peeling paint, natural light photography.
中文說明:銀色虎斑緬因貓坐在窗台,長針毛與鬍鬚一根根分明,耳尖毛被午後陽光逆光,綠眼虹膜紋理可見,木窗框油漆剝落,自然光。
A06. 旗袍刺繡
Detail shot of a silk qipao collar with hand-embroidered peonies and a golden phoenix, satin-stitch threads catching the light, a fabric knot button, faint weave of the silk visible, warm tungsten light in a tailor's shop, photorealistic textile photography.
中文說明:絲質旗袍領口特寫,手繡牡丹與金鳳凰,緞面繡線反光,布製盤扣,絲綢織紋隱約可見,裁縫店暖色鎢絲燈,寫實織品攝影。
A07. 珊瑚微距
Underwater macro photograph of a brain coral and a cluster of orange tube coral polyps, tiny translucent shrimp on the coral, particles floating in the water column, strobe-lit against a deep blue background, sharp focus on the coral texture.
中文說明:水下微距:腦珊瑚與一叢橘色管珊瑚水螅體,珊瑚上有半透明小蝦,水中懸浮微粒,閃燈打亮、深藍背景,珊瑚紋理清晰。
A08. 酸種麵包
A sliced sourdough loaf on a wooden board, blistered caramelised crust with flour dust and a scored ear, open crumb with glossy irregular holes, crumbs scattered, morning side light, food photography, 50mm lens, extremely detailed.
中文說明:切開的酸種麵包放在木板上,起泡焦糖化外皮、灑麵粉、割紋耳朵,內部大小不一的亮面氣孔,散落麵包屑,晨光側光,食物攝影 50mm。
B. 文字與版面
B01. 夜市攤位招牌
A busy Taiwanese night market food stall at dusk, a glowing yellow signboard above it reads "基隆廟口鹽酥雞" in bold black Traditional Chinese characters, a smaller red vertical banner on the left reads "現炸 每份60元", steam rising from a fryer, hanging light bulbs, photorealistic street photography.
中文說明:黃昏的台灣夜市小吃攤,上方發亮黃色招牌寫「基隆廟口鹽酥雞」,左側紅色直式小旗寫「現炸 每份60元」,油鍋蒸氣、吊燈泡,寫實街拍。埋字:廟、鹽。
B02. 手搖飲菜單
A hand-lettered chalkboard menu of a Taiwanese bubble tea shop, white and pastel chalk on black, title "手搖飲菜單" at the top, four items listed in Traditional Chinese: "珍珠鮮奶茶 65", "烏龍鐵觀音 45", "檸檬冬瓜露 50", "黑糖波霸鮮奶 70", small chalk drawings of cups between the lines, shallow depth of field, cafe interior.
中文說明:手搖飲店的手寫黑板菜單,白色與粉彩粉筆,標題「手搖飲菜單」,四個品項與價格:珍珠鮮奶茶 65、烏龍鐵觀音 45、檸檬冬瓜露 50、黑糖波霸鮮奶 70,行間有粉筆畫的杯子,淺景深。
B03. 廟宇對聯
Entrance of a traditional Taiwanese temple, red wooden pillars on both sides each carrying a vertical gold-lettered couplet in Traditional Chinese calligraphy, the right pillar reads "龍飛鳳舞護蒼生" and the left pillar reads "鬱鬱蒼松迎瑞氣", above the door a horizontal plaque reads "慈祐宮", incense smoke, warm lantern light, photorealistic.
中文說明:台灣傳統廟宇入口,兩側紅柱各掛一副金字直式對聯,右聯「龍飛鳳舞護蒼生」、左聯「鬱鬱蒼松迎瑞氣」,門上橫匾「慈祐宮」,香煙、暖色燈籠光。埋字:龍、鬱。
B04. 車站看板
A railway station departure board in Taiwan, dark panel with amber LED text in Traditional Chinese, column headers "車次 目的地 時間 月台", three rows: "123 花蓮 08:15 2", "411 臺灣高雄 08:42 1", "507 龜山 09:03 3", a small red line at the bottom reads "請讓座給需要的人", eye-level shot, station hall slightly out of focus behind.
中文說明:台灣火車站的發車看板,深色面板琥珀色 LED 字,欄位標題「車次 目的地 時間 月台」,三列:123 花蓮 08:15 2、411 臺灣高雄 08:42 1、507 龜山 09:03 3,底部一行紅字「請讓座給需要的人」。埋字:臺、灣、龜、讓。
B05. 牛肉麵店價目表
Interior of an old Taiwanese noodle shop, a wall-mounted wooden price board with hand-painted Traditional Chinese text: title "老張牛肉麵", items "紅燒牛肉麵 180", "清燉牛肉麵 180", "乾麵 60", "滷蛋 15", a paper note taped beside it reads "今日公休", fluorescent light, tiled walls, documentary photo.
中文說明:老牛肉麵店內部,牆上手繪木製價目板:標題「老張牛肉麵」,品項紅燒牛肉麵 180、清燉牛肉麵 180、乾麵 60、滷蛋 15,旁邊貼一張紙條「今日公休」,日光燈、磁磚牆,紀實照。埋字:麵。
B06. 春聯書法
A red spring festival couplet paper on a wooden door, brush calligraphy in black ink reading "籲天賜福平安至" vertically, with a golden diamond-shaped paper above showing the single character "疊" rotated 45 degrees, ink texture and paper grain visible, soft daylight, close-up.
中文說明:木門上的紅色春聯,黑墨毛筆直書「籲天賜福平安至」,上方菱形金紙寫單一個「疊」字旋轉 45 度,墨跡與紙紋可見,柔和日光特寫。埋字:籲、疊。
B07. 漫畫對話框
A single manga-style panel in black and white ink, a girl in a school uniform holding an umbrella in the rain, a speech bubble above her reads "這場雨還要下多久啊?" in Traditional Chinese, a small sound-effect text "嘩啦嘩啦" drawn near the umbrella edge, screentone shading, clean line art.
中文說明:黑白漫畫單格,穿制服的女孩雨中撐傘,對話框寫「這場雨還要下多久啊?」,傘邊有狀聲字「嘩啦嘩啦」,網點陰影、乾淨線稿。
B08. 超商促銷海報
A Taiwanese convenience store promotional poster, bright flat graphic design, large headline "鹽味海苔飯糰 第二件八折" in bold Traditional Chinese, a subtitle "活動期間 9/10 - 9/30", a product photo of a triangular rice ball in the centre, small footer text "限量供應 售完為止", clean vector style.
中文說明:台灣超商促銷海報,平面設計,大標題「鹽味海苔飯糰 第二件八折」,副標「活動期間 9/10 - 9/30」,中央三角飯糰產品照,底部小字「限量供應 售完為止」。埋字:鹽、糰。
B09. 打字機信件
A sheet of cream paper in a vintage typewriter, the typed text reads exactly: "Dear Margaret, The harbour was quiet this morning. Fog sat on the water until nine, and the fishing boats stayed tied to the pier. I walked to the lighthouse and thought about the summer you visited. Write soon. Yours, Thomas" in monospaced typewriter font, slight ink irregularities, warm desk lamp light, close-up.
中文說明:老式打字機上的奶油色紙,打字內容逐字指定(一封給 Margaret 的短信,約 60 字),等寬打字機字體、墨色不均、暖色檯燈。測整段英文長文。
B10. 早餐麥片盒
Front of a retro cereal box, four text regions: brand name "MOONFLAKES" in large yellow 3D letters at the top, tagline "Crunch that lasts till lunch" in a red ribbon below it, a badge on the right reading "NEW! 30% more oats", and a bottom strip reading "NET WT 18 OZ (510g)", a cartoon fox mascot holding a spoon, 1970s packaging design, flat front view.
中文說明:復古麥片盒正面,四個文字區:頂部黃色 3D 大字品牌「MOONFLAKES」、紅色緞帶標語「Crunch that lasts till lunch」、右側徽章「NEW! 30% more oats」、底部「NET WT 18 OZ (510g)」,卡通狐狸吉祥物,1970 年代包裝風格。測多區域版面。
B11. 黑板公式推導
A university lecture chalkboard, white chalk handwriting: the heading "Kinetic Energy", then the lines "W = ∫ F dx", "F = m dv/dt", "W = ∫ m v dv", "W = ½ m v²" written in a column, a small boxed note at the lower right reads "Exam: Friday 10am", chalk dust, wooden chalk tray, slightly low angle.
中文說明:大學講堂黑板,白粉筆手寫:標題「Kinetic Energy」,四行公式由 W = ∫ F dx 推到 W = ½ m v²,右下角方框註記「Exam: Friday 10am」,粉筆灰、木製粉筆槽,微仰角。
B12. 機場三語指標
A modern airport wayfinding sign hanging from the ceiling, dark blue panel with white text in three languages stacked: "Baggage Claim", "行李提領", "手荷物受取所", an arrow pointing left, next to it a yellow panel reading "Exit 出口" with an arrow pointing right, glass and steel terminal behind, photorealistic.
中文說明:現代機場吊掛指標,深藍底白字三語疊排:Baggage Claim/行李提領/手荷物受取所,左箭頭;旁邊黃色板「Exit 出口」右箭頭,玻璃鋼構航廈背景。測中英日混排。
B13. 台灣茶區資訊圖
A clean infographic poster titled "Taiwan Tea Regions 台灣茶區" with a simplified island map in the centre and five callout labels connected by thin lines: "Alishan 阿里山 High Mountain Oolong", "Muzha 木柵 Tieguanyin", "Sun Moon Lake 日月潭 Black Tea", "Wenshan 文山 Baozhong", "Lugu 鹿谷 Dong Ding", each with a small tea leaf icon, sage green and cream palette, sans-serif typography.
中文說明:資訊圖海報,標題「Taiwan Tea Regions 台灣茶區」,中央簡化台灣地圖,五個標籤以細線連出:阿里山高山烏龍、木柵鐵觀音、日月潭紅茶、文山包種、鹿谷凍頂,各附小茶葉圖示,鼠尾草綠與米色配色。埋字:柵。
C. 編輯與多參考圖合成
參考圖用 Krea2 Turbo 生成,prompt 如下(先生出這些圖,再拿它們當參考):
| 參考圖 | 用途 | Prompt |
ref_e1_headphones | 戴大耳機、灰帽 T 的男人在咖啡館窗邊(C01 的編輯來源) | Waist-up portrait of a man in his thirties wearing large black over-ear headphones and a grey hoodie, sitting in a bright cafe by a window, neutral expression, photorealistic, 50mm. |
ref_e2_bakery_sign | 木招牌寫「MORNING LOAF」的麵包店店面(C02) | Storefront of a small bakery with a wooden signboard above the door reading "MORNING LOAF" in white serif letters, croissants in the window, brick facade, morning light, straight-on shot. |
ref_e3_living_room | 北歐風客廳、淺灰布沙發(C03) | A Scandinavian living room with a light grey fabric sofa, a round oak coffee table, a potted fiddle-leaf fig, white walls, large window with sheer curtains, daylight, interior photography. |
ref_e4_street_day | 歐洲老城正午街道(C04) | A quiet European old-town street at midday, cobblestones, pastel facades, a bicycle leaning on a wall, a cafe with two outdoor tables, clear sky, photorealistic. |
ref_e5_bottle_label | 標籤寫「CALM SERUM 30 ml」的磨砂玻璃瓶(C05) | A frosted glass skincare bottle on a marble surface, a minimal white label with the text "CALM SERUM" and below it "30 ml", soft studio lighting, product photography. |
ref_s1_girl_front | 紅辮子、黃雨衣、玳瑁圓眼鏡、綠雨鞋的女孩正面(C06) | Full-body character sheet style photo of a girl about ten years old with a thick red braid over one shoulder, a bright yellow raincoat, round tortoiseshell glasses and green rubber boots, standing facing the camera on a plain light grey background, soft even lighting. |
ref_s1_girl_side | 同一女孩的四分之三側面(C06) | Three-quarter side view of the same girl about ten years old with a thick red braid, bright yellow raincoat, round tortoiseshell glasses and green rubber boots, plain light grey background, soft even lighting. |
ref_s2_corgi | 柯基(C07) | A Pembroke Welsh corgi with a white blaze and one folded ear sitting on a plain white background, studio pet photography. |
ref_s2_mug | 燕麥色斑點釉、鈷藍杯緣的手拉坯馬克杯(C07) | A hand-thrown ceramic mug with a speckled oatmeal glaze and a cobalt blue rim, plain white background, product photo. |
ref_s3_man_front | 鬍子花白、光頭、左眉有疤的男人(C08) | Head-and-shoulders photo of a man in his forties with a salt-and-pepper beard, shaved head and a small scar over the left eyebrow, plain grey background, neutral expression. |
ref_s3_jacket | 皮製飛行夾克、羊毛領、左袖有徽章(C08) | A vintage brown leather bomber jacket with a shearling collar and a patch on the left sleeve, hanging on a plain white background, product photo. |
ref_s3_diner | 1950 年代美式餐館夜景(C08) | Interior of an empty 1950s American diner at night, red vinyl booths, chrome counter, neon sign glowing pink, checkered floor, photorealistic. |
ref_s4_robot_front | 青綠配橘色的圓身維修機器人正面(C09) | A small round-bodied maintenance robot, matte teal shell with orange accents, a single large circular lens for an eye, two stubby arms, tank treads, front view on a plain white background, 3D render. |
ref_s4_robot_back | 同一機器人背面(C09) | Back view of the same small round-bodied maintenance robot, matte teal shell with orange accents, a vent grille on the back, tank treads, plain white background, 3D render. |
C01. 移除耳機
Remove the headphones from the man. Keep everything else exactly the same.
參考圖:ref_e1_headphones
中文說明:指令:把男人的耳機拿掉,其他全部保持不變。
C02. 招牌換字
Change the signboard text "MORNING LOAF" to "晨光麵包坊" in the same white serif style. Keep the storefront, lighting and everything else unchanged.
參考圖:ref_e2_bakery_sign
中文說明:指令:把招牌上的「MORNING LOAF」換成「晨光麵包坊」,同樣的白色襯線風格,店面、光線與其他一切不變。
C03. 換沙發加貓
Change the sofa to emerald green velvet and add a sleeping orange tabby cat on it. Keep the room layout, table, plant and lighting the same.
參考圖:ref_e3_living_room
中文說明:指令:沙發換成祖母綠絲絨,加一隻睡著的橘虎斑貓,房間格局、桌子、植物、光線不變。
C04. 白天變雨夜
Turn this scene into a rainy night: wet reflective cobblestones, warm light from the cafe windows, street lamps on, light rain visible. Keep the buildings, bicycle and tables in place.
參考圖:ref_e4_street_day
中文說明:指令:把場景改成雨夜,濕亮的鵝卵石、咖啡館窗戶透出暖光、路燈亮起、可見細雨,建築、腳踏車與桌子位置不變。
C05. 標籤換字
Replace the label text with "舒緩精華液" on the first line and "30 毫升" on the second line, same minimal typography. Keep the bottle, marble and lighting unchanged.
參考圖:ref_e5_bottle_label
中文說明:指令:標籤第一行改成「舒緩精華液」、第二行改成「30 毫升」,同樣極簡字體,瓶子、大理石與光線不變。
C06. 女孩到夜市
The girl from the reference images is standing in a rainy Taipei night market, holding a paper bag of steamed buns, neon signs reflecting in puddles, photorealistic, medium shot.
參考圖:ref_s1_girl_front、ref_s1_girl_side
中文說明:兩張參考圖(正面、側面)裡的女孩站在雨夜的台北夜市,手拿一袋包子紙袋,霓虹倒映在水窪,寫實中景。
C07. 柯基與馬克杯
The corgi from the first reference sits at a wooden breakfast table next to the ceramic mug from the second reference, morning sunlight, a window with plants behind, photorealistic.
參考圖:ref_s2_corgi、ref_s2_mug
中文說明:第一張參考圖的柯基坐在木製早餐桌旁,旁邊放第二張參考圖的陶瓷馬克杯,晨光、窗邊有植物,寫實。
C08. 人、夾克、餐館
The man from the first reference, wearing the leather jacket from the second reference, sits alone in a booth of the diner from the third reference, drinking coffee, cinematic lighting, medium shot.
參考圖:ref_s3_man_front、ref_s3_jacket、ref_s3_diner
中文說明:第一張參考圖的男人穿著第二張參考圖的皮夾克,獨自坐在第三張參考圖的餐館卡座喝咖啡,電影感打光,中景。
C09. 機器人到海邊
The robot from the reference images rolls along a wet sandy beach at sunset, seagulls overhead, its lens reflecting the orange sky, low-angle shot, photorealistic 3D render.
參考圖:ref_s4_robot_front、ref_s4_robot_back
中文說明:參考圖裡的機器人在日落的濕沙灘上前進,海鷗掠過,鏡片反射橘色天空,低角度,寫實 3D 渲染。
D. 分鏡與鏡位
D01. 燈塔守衛四格分鏡
A four-panel storyboard in a 2x2 grid with thin white gutters, consistent painterly style and the same character in every panel: an old lighthouse keeper with a white beard and a navy peacoat. Panel 1 (top left): he climbs the spiral stairs at dusk holding a lantern. Panel 2 (top right): he polishes the huge lens as the storm begins outside. Panel 3 (bottom left): wide shot of the lighthouse beam cutting through rain over a struggling fishing boat. Panel 4 (bottom right): morning, he waves from the gallery to the boat safely in the harbour. Each panel has a small number in its corner.
中文說明:2×2 四格分鏡、白色細間隔,油畫風格,每格同一個白鬍子、海軍呢大衣的老燈塔守衛:黃昏提燈爬螺旋梯/暴風雨中擦拭巨大透鏡/燈塔光束穿過雨幕照向掙扎的漁船/清晨在迴廊向安全進港的船揮手,每格角落有編號。
D02. 機器人烘焙四格分鏡
A four-panel storyboard in a horizontal strip, flat cartoon style, the same small teal robot in every panel. Panel 1: the robot reads a recipe card in a sunny kitchen. Panel 2: flour explodes everywhere as it mixes dough too fast. Panel 3: it stares nervously at a smoking oven. Panel 4: it proudly presents a slightly burnt but heart-shaped loaf to a smiling little girl. Thin black panel borders, numbered panels.
中文說明:橫向四格分鏡、扁平卡通風,每格同一個小型青綠色機器人:陽光廚房讀食譜卡/攪拌太快麵粉爆開/緊張盯著冒煙的烤箱/自豪地把有點焦的心形麵包遞給笑著的小女孩,細黑框、格內編號。
D10–D24. 15 種鏡位
固定主體(每一題結尾都接這句):
a woman street violinist in a long burgundy coat playing under a stone archway in a rainy old European city at night, warm lamplight, wet reflective cobblestones, photorealistic
每題只換開頭的鏡位描述,15 張共用同一個 seed:
| 鏡位 | 描述 |
| extreme wide | extreme wide shot, the subject tiny within the whole square and surrounding buildings |
| wide | wide shot, full archway and street visible, subject small but clear |
| full | full shot, the subject from head to toe filling most of the frame height |
| medium full | medium full shot, framed from the knees up |
| medium | medium shot, framed from the waist up |
| medium closeup | medium close-up, framed from the chest up |
| closeup | close-up of her face and the violin's scroll |
| extreme closeup | extreme close-up of her fingers on the violin strings |
| low angle | low-angle shot from near the ground looking up at her |
| high angle | high-angle shot looking down at her from a balcony |
| birds eye | bird's-eye view directly from above, the archway and her small figure below |
| worms eye | worm's-eye view from the cobblestones straight up, her silhouette against the arch |
| dutch angle | Dutch angle, the camera tilted about 20 degrees |
| over shoulder | over-the-shoulder shot from behind a listener in the foreground, the violinist in focus beyond |
| pov | first-person point-of-view shot as if the viewer is the violinist, looking down the violin at the crowd |
中文說明:雨夜歐洲老城石拱門下、穿酒紅長大衣的女街頭小提琴手。15 種鏡位依序是極遠景、遠景、全身、中全身、中景、中特寫、特寫、極特寫、仰角、俯角、鳥瞰、蟲視、荷蘭角、過肩、第一人稱。
E. 需要背景知識的短句,與 Agent 改寫版
改寫用的是官方 prompt_agent.py 的系統提示詞(一字未改),語言模型換成本機跑的 Qwen 27B。Agent 回傳的除了改寫後的 prompt,還有它解析出的隱含知識,一併列出。
E01. 鹿柴竹簡
原始 prompt:
王維的《鹿柴》刻在一片竹簡上
Agent 解析出的知識:解析了王维《鹿柴》的全诗内容:'空山不見人,但聞人語響。返景入深林,復照青苔上。'。确定了竹简的材质特征(淡黄色、纵向纹理)和适合古典诗词的书法字体(篆书)。
Agent 改寫版:
A close-up, high-angle shot of a single, weathered bamboo slip lying diagonally across a rough, mossy stone surface. The bamboo is pale yellow with natural vertical grain and slight darkening at the edges, showing signs of age and exposure to elements. Carved deeply into the surface in traditional Chinese seal script (Zhuanshu) are the exact characters: "空山不見人,但聞人語響。返景入深林,復照青苔上。" The carving is precise, with dark ink or natural shadow filling the grooves to make the text legible against the light bamboo background. A single dry leaf rests near the top right corner of the slip, adding a touch of autumnal decay. The lighting is soft, diffused natural light, suggesting a shaded forest clearing, creating gentle highlights on the bamboo texture and soft shadows beneath the slip. The background is a blurred, out-of-focus green forest floor with hints of ferns and scattered leaves, enhancing the depth of field. The style is photorealistic, with sharp focus on the text and bamboo texture, evoking a serene, classical, and contemplative atmosphere consistent with Wang Wei's poetry.
中文說明:原始 prompt 只有一句「王維的《鹿柴》刻在一片竹簡上」,模型得自己知道詩句。
E02. 牛頓展區
原始 prompt:
a science museum exhibit about the person who formulated the three laws of motion
Agent 解析出的知識:用户提到的'制定运动三大定律的人'指的是艾萨克·牛顿(Isaac Newton)。牛顿三大定律分别是:第一定律(惯性定律)、第二定律(F=ma)、第三定律(作用与反作用定律)。我将这些抽象的物理概念转化为具体的视觉元素,如炮弹轨迹、天平上的不同质量物体、以及滑冰者互相推开的场景。同时,我加入了牛顿的标志性物品,如苹果和《自然哲学的数学原理》,以增强展品的教育意义和历史准确性。
Agent 改寫版:
A high-resolution photorealistic wide shot of a modern science museum exhibit dedicated to Sir Isaac Newton. The central focus is a life-sized, highly detailed bronze statue of Isaac Newton, dressed in 17th-century academic robes, seated on a wooden chair with a contemplative expression, holding an open book in one hand and a polished apple in the other. Behind the statue, a large curved digital screen displays a dynamic, animated visualization of Newton's three laws of motion: a cannonball trajectory representing the first law (inertia), a scale with two weights of different masses illustrating the second law (force equals mass times acceleration), and two ice skaters pushing off from each other to demonstrate the third law (action and reaction). To the left of the statue, a glass display case contains a replica of Newton's reflecting telescope and a leather-bound copy of his 'Principia Mathematica', with the title clearly visible on the spine. To the right, an interactive pedestal features a glowing hologram of a pendulum swinging, labeled with the equation 'F = ma' in bright white sans-serif text. The lighting is dramatic and focused, with soft spotlights highlighting the statue and the holographic elements, while the surrounding museum floor is a polished, dark grey stone. The atmosphere is educational, immersive, and futuristic, with a color palette dominated by deep blues, silvers, and warm bronze tones. The composition uses a low-angle perspective to make the statue appear grand and authoritative, with a shallow depth of field that keeps the background exhibits slightly blurred to draw attention to the central figure and the digital screen.
中文說明:原始 prompt 沒有點名牛頓,只說「提出運動三定律的那個人」的博物館展區。
E03. 台北跨年煙火
原始 prompt:
台北101跨年煙火,從象山看過去
Agent 解析出的知識:象山 (Elephant Hill) 是台北著名的观景台,位于台北101的南面/西南面,从象山看101,101位于远处,前景有山坡植被和路灯剪影。台北101跨年烟火通常伴随大楼LED灯光秀,颜色多为喜庆的红、金、蓝。
Agent 改寫版:
A breathtaking cinematic wide-angle night view of the Taipei 101 skyscraper, as seen from the elevated vantage point of Elephant Hill. The massive, tiered tower stands prominently in the center-left of the frame, its sleek green glass facade illuminated by millions of LED lights in vibrant red, gold, and blue patterns. A spectacular, chaotic burst of golden and silver fireworks explodes directly above the building's spire, creating a dazzling crown of light that reflects on the wet city streets below. The foreground features the dark, silhouetted, undulating ridges of Elephant Hill with scattered city lights, providing a natural frame. The background shows the dense, hazy urban skyline of Taipei stretching into the distance under a deep indigo sky. The image is a photorealistic long-exposure shot, 24mm lens, low angle looking slightly up, capturing the grandeur and scale of the New Year's Eve celebration with high contrast and luminous clarity.
中文說明:原始 prompt:台北101跨年煙火,從象山看過去。模型得知道象山與 101 的相對位置。
E04. 颱風警報 App
原始 prompt:
a weather app screen showing a typhoon warning for Taiwan
Agent 解析出的知識:无
Agent 改寫版:
A high-fidelity UI design mockup of a weather application screen on a modern smartphone, displaying a severe typhoon warning for Taiwan. The top half of the screen features a vivid, stylized map of the island of Taiwan with a prominent, swirling dark red and orange typhoon symbol positioned over the central region, indicating an approaching storm. Below the map, a large, bold, sans-serif text in white on a dark semi-transparent banner reads 'Severe Typhoon Warning: Taiwan'. The background of the app is a deep, stormy blue with subtle, animated rain streaks and lightning flashes. The UI elements include a temperature display showing 28°C, a wind speed indicator at 90 km/h, and a humidity level at 85%, all in crisp, white, modern fonts. The overall aesthetic is clean, professional, and high-tech, with a focus on clarity and urgency. The phone is shown in a close-up, centered composition, with a slight tilt to add depth, and the screen is the focal point, sharp and in focus, while the surrounding environment is a soft, out-of-focus indoor setting with warm, ambient lighting. 85mm lens, shallow depth of field, photorealistic rendering, 4K resolution.
中文說明:原始 prompt:顯示台灣颱風警報的天氣 App 畫面。測 UI 版面與文字。
E05. 莫比烏斯鐵軌
原始 prompt:
a Möbius strip made of railway track with a toy train on it
Agent 解析出的知識:莫比乌斯环(Möbius strip):一种只有一个面和一条边界的拓扑结构,由一条带子扭转180度后首尾相连形成。铁轨:通常由钢轨、枕木和道砟组成,这里简化为连续的钢轨结构以形成环状。玩具火车:小型模型火车,通常具有鲜艳颜色和细节结构。
Agent 改寫版:
A surreal, hyper-realistic 3D render of a continuous Möbius strip formed entirely by a single, weathered steel railway track, floating in a vast, empty void with a soft, gradient gray background. The track twists seamlessly, creating a non-orientable surface where the inner and outer edges merge, defying standard geometry. A small, detailed vintage toy train, painted in vibrant red and cream with brass accents and visible rivets, is positioned on the upper curve of the strip, moving forward with its headlight casting a warm, soft beam onto the tracks ahead. The metal of the rails shows realistic reflections and subtle rust textures, emphasizing the industrial material. The composition is centered, with a slightly low-angle perspective to emphasize the impossible loop. Lighting is studio-quality, soft and diffused, highlighting the curves and the smooth transition of the twist. High-resolution, 8k, sharp focus, cinematic depth of field, clean minimalist aesthetic.
中文說明:原始 prompt:用鐵軌做成的莫比烏斯環,上面有一輛玩具火車。測拓撲結構的理解。
E06. 手沖咖啡參數卡
原始 prompt:
一杯手沖咖啡從正上方看,旁邊放一張沖煮參數卡
Agent 解析出的知識:手冲咖啡的典型视觉特征(深色液体、油脂、陶瓷杯);常见的手冲参数标准(90-95°C水温,1:15-1:17粉水比,中细研磨,2-3分钟萃取时间);平铺摄影的构图逻辑(主体居中,辅助物在侧)。
Agent 改寫版:
A top-down, flat-lay photorealistic photograph featuring a ceramic cup of freshly brewed pour-over coffee, centered in the foreground, filled with dark amber liquid showing subtle surface crema and gentle ripples. To the right of the cup, placed on a light oak wooden table, is a small, rectangular white paper card titled 'BREWING PARAMETERS' in a clean, black sans-serif font. The card lists specific details in a neat, legible layout: 'Water Temp: 92°C', 'Ratio: 1:15', 'Grind: Medium-Fine', and 'Time: 2:30'. The lighting is soft, natural daylight coming from the top left, casting soft, realistic shadows that emphasize the texture of the ceramic and the grain of the wood. The color palette is warm and earthy, focusing on beige, brown, and white tones. Shot with a 50mm macro lens, high resolution, sharp focus on the text and coffee surface, shallow depth of field blurring the distant edges of the table slightly to draw attention to the central elements.
中文說明:原始 prompt:一杯手沖咖啡從正上方看,旁邊放一張沖煮參數卡。模型得自己補出參數卡上該寫什麼。
E07. 孤獨
原始 prompt:
loneliness
Agent 解析出的知識:无
Agent 改寫版:
A photorealistic cinematic medium shot captures a solitary figure, a man in a worn, dark wool overcoat with a high collar, standing motionless on a vast, empty concrete pier extending into a misty, gray ocean. He is positioned slightly to the left of the frame, facing the horizon with his back partially turned to the camera, head bowed in a posture of deep contemplation and isolation. The background features a dense, low-hanging fog that obscures the water's edge and distant land, creating a seamless blend of sky and sea in muted tones of slate gray and pale blue. The lighting is soft, diffuse, and overcast, casting no harsh shadows and evoking a cold, melancholic atmosphere. The composition utilizes significant negative space on the right side to emphasize the emptiness surrounding the subject. Shot on a 50mm lens with a shallow depth of field, keeping the man sharp while the foggy horizon softens into a dreamlike blur. The color palette is desaturated, dominated by cool grays, blues, and blacks, enhancing the feeling of solitude and quiet sadness.
中文說明:原始 prompt 只有一個字:loneliness。測抽象概念具象化。
E08. 汝窯天青釉
原始 prompt:
北宋汝窯天青釉瓷碗的特寫
Agent 解析出的知識:北宋汝窑特征:天青釉色(Sky-blue/Turquoise glaze),玛瑙入釉带来的玉质感,蟹爪开片(Ice crack pattern),紫口铁足(Purple mouth, iron foot),支钉烧制(Tooth-pin support,虽特写可能看不到底部支钉,但足部颜色特征需体现)。
Agent 改寫版:
Extreme close-up of a Northern Song Dynasty Ru kiln celadon bowl, showcasing its iconic sky-blue glaze with a subtle, misty turquoise tint. The surface features the characteristic 'ice crack' pattern, consisting of fine, web-like fissures that are barely visible under raking light. The rim is thin and unglazed, revealing a dark, iron-rich brown edge known as 'purple mouth.' The base is covered in a thick layer of iron-rich 'temple earth' foot ring, appearing dark brown and rough in contrast to the smooth, matte, and creamy texture of the glaze. Soft, diffused studio lighting highlights the translucent depth of the glaze and the delicate, organic curves of the small, elegant bowl. The background is a dark, neutral charcoal grey to enhance the pale blue tones. Photorealistic, high-resolution product photography, 100mm macro lens, shallow depth of field, sharp focus on the glaze texture, 8k.
中文說明:原始 prompt:北宋汝窯天青釉瓷碗的特寫。測模型認不認得汝窯的特徵(天青、開片、紫口鐵足)。
F. 長寬比
A panoramic view of Jiufen old street at blue hour, tiered red lanterns along the stone stairs, tea houses with glowing windows, misty mountains and the sea in the distance, a few umbrellas, photorealistic travel photography
三種尺寸各在句尾加一句構圖提示:2560×1440 加「wide cinematic landscape composition」、1440×2560 加「tall vertical composition following the stairs upward」、3104×1312 加「ultra-wide panoramic composition」。
中文說明:藍調時刻的九份老街全景,石階兩側層層紅燈籠、茶樓窗戶透光、遠處霧山與海、幾把雨傘,寫實旅遊攝影。
相關文章
- HiDream-O1 實測:官方說的六件事,本機驗證對了幾件 — 這套題目的成績與全部對照圖
- 繁體中文 AI 生圖測試 Prompt:53 組全公開 — 專測繁體中文的另一套題庫
- SenseNova 官方 Prompt 增強實測:AI 生圖的亂碼不是隨機的 — 另一家的 prompt 改寫層,結論相同




