01
任務與我的職責
需求是使用 IP 角色生成音樂短片,用於音樂作品推廣。我參與優化已有工作流,工作集中在提示詞、分鏡與生成效果。
重點處理兩個問題:角色在不同鏡頭中的一致性,以及畫面切換與音樂節奏的匹配。
02
圍繞重拍安排鏡頭
在流程中加入對音訊 BPM 與節奏型的分析要求,將鏡頭切換安排在重拍位置,作為畫面卡點的依據。
節奏分析與鏡頭安排是生成約束,最終仍需要檢查實際影片中的音畫表現。
03
先形成角色分鏡,再生成影片
採用先生成包含 IP 角色的分鏡圖、再據此生成影片的策略,為不同片段提供更明確的視覺參考。
這一調整針對角色一致性問題,讓影片生成階段有具體的角色畫面可以依循。
04
把參考作品轉化為提示詞
針對音畫匹配回饋,我參考近五年的 K-pop MV,拆解其中的運鏡與剪輯方式,將示例整理進系統提示詞,幫助工作流組織鏡頭語言。
團隊完成效果後交給後端部署;我的貢獻是已有工作流中的提示詞與內容效果優化。
我將不同的音樂影片視覺參考整理為七組場景提示詞,分別約束構圖、空間、配色、光照與不希望出現的元素。下面保留提供的英文原文;藝人名稱作為風格分類標籤。
把視覺判斷寫成具體條件
例如,TWICE 條目把傢俱安排到邊緣、保留中央空地,並指定粉色、薄荷綠和檸檬黃;STRAY KIDS 條目要求平視、水平構圖和中央無雜物;NEWJEANS 條目則將場景、光照、相機參數、材質與負向提示分開組織。
這些片段體現了場景設計的約束方式。各風格的要求並不完全相同,例如 ATEEZ 條目保留灰塵與碎片來形成廢棄感。
TWICE - 元氣流行展開提示詞原文
Cinematic wide-angle shot of a vibrant, hyper-aesthetic K-pop music video set. A massive, sun-drenched penthouse hall with a vast, empty polished wooden dance floor in the center. All furniture and stylish colorful cardboard boxes are neatly stacked at the far edges. Large floor-to-ceiling windows showing an overexposed skyline. Palette: Pastel pink, mint green, and lemon yellow (candy colors). Modern architectural curves on the ceiling with glossy textures. High-key natural lighting with strong lens flares. Pop-art aesthetic, surreal and clean. (Negative: humans, people, cluttered center, dark shadows, night, messy floor, cartoon)
BLACKPINK - 科技未來展開提示詞原文
Symmetrical, eye-level wide shot of a futuristic K-pop urban stage at night. A vast, expansive empty asphalt center designed for dance performance, reflecting vibrant neon lights. The powder-blue convertible and teal-green sedan are neatly parked at the extreme left and right foreground edges, framing the scene. Background features a highly detailed building with massive glowing neon "CISUMVERSE" and "STARNEST" signs, enhanced with intricate LED light tubes and holographic elements. Straight-on horizontal perspective, no tilt. High-contrast cinematic lighting with a palette of electric cyan and hot pink. Wet ground with crystal clear reflections. Volumetric fog and professional stage lighting. High-gloss textures, exquisite urban aesthetic. (Negative: people, humans, Dutch angle, slanted view, cluttered center, debris, barricades in middle, cartoon, low resolution)
STRAY KIDS - 工業廢土展開提示詞原文
Symmetrical eye-level wide shot of a massive, hyper-exquisite industrial warehouse transformed into a high-budget K-pop stage. A vast, perfectly empty performance area in the center with a polished, reflective wet concrete floor, ideal for a five-person dance routine. The center is completely clear of debris. Framing the stage are giant, glowing industrial turbine fans and heavy metal chains hanging from the high ceiling. Concrete pillars are accented with vertical vibrant orange and electric blue neon light tubes. Harsh, dramatic cinematic floodlighting with volumetric fog and deep shadows. Gritty but luxurious dystopian aesthetic. Perfectly straight horizontal perspective, centered composition. (Negative: tilted, Dutch angle, slanted, debris in center, scaffolding blocking center, people, humans, warm sunlight, bright happy colors, low resolution)
IVE - 奢華大片展開提示詞原文
Symmetrical eye-level wide shot of an ultra-luxurious K-pop runway set on a vast, mirror-like black reflective airfield at night. A perfectly straight horizontal perspective with a massive empty central dance floor. The road is flanked by towering high-tech scaffolding structures and immense vertical LED screens glowing with elegant shimmering gold and brilliant white abstract graphics. Multiple rows of powerful floor-mounted stage lights shoot intense, sharp golden beams and pure white light across the scene, creating a majestic golden light tunnel effect. The ground is wet and highly reflective, mirroring the luxurious golden radiance and white sparkles. Neoclassical cyberpunk aesthetic, high-gloss gold textures, cinematic 8k resolution. (Negative: cyan, pink, blue, magenta, purple, bird's-eye view, high angle, tilted view, humans, people, daylight, sun, rustic, grassy, dirt road, messy, cartoon, low resolution)
ATEEZ - 末世史詩展開提示詞原文
Cinematic wide-angle interior of a grand abandoned ballroom in decay. Tattered fabric hanging from ceiling. Floor covered in thick gray dust and debris. Strong dramatic volumetric lighting (god rays) streaming from left windows. Deep murky shadows on right. Post-apocalyptic luxury, haunted mansion aesthetic. (Negative: people, clean, renovated, bright artificial lighting, vibrant colors)
BTS - 詩意敍事展開提示詞原文
Symmetrical eye-level shot with a 35mm lens, showcasing a vast and flat sandy desert plateau at the peak of a breathtaking sunset. The foreground is an expansive, even area of fine-grained sand, serving as a perfectly flat stage for a dance performance. In the far background, majestic rolling sand dunes and hazy mountain silhouettes stretch across the horizon. The sky is a magnificent canvas of deep crimson, fiery orange, and brilliant violet afterglow. Cinematic high-budget K-pop music video aesthetic, poetic and epic atmosphere. Perfectly straight horizontal perspective, centered composition. (Negative: peaked dunes in center, sharp sand ridges in foreground, chairs, school furniture, reflection, upside down, people, humans, clouds, rain, vegetation, dark, gloomy, noise, cluttered, low resolution)
NEWJEANS - 復古文藝展開提示詞原文
[Composition/Location] Wide-angle eye-level shot looking toward an urban railroad crossing in a densely populated East Asian city. In the mid-ground, a white and yellow commuter train is blurred as it speeds across the frame horizontally. The foreground features a gray asphalt road with distinct yellow double lines. Overhead, two massive yellow industrial steel truss structures with "4.2M" height limit signs and electronic traffic signals span the road. Background shows high-rise residential apartment buildings with visible air conditioning units and a lush green hillside in the far distance. [Lighting Logic] Bright, high-contrast natural daylight. The scene is illuminated by direct overhead sun, creating sharp, short shadows. The light reflects off the metallic surfaces of the train and the industrial yellow trusses, giving the entire environment a clean, slightly overexposed urban summer vibe. [Camera Parameters] Shot on a high-end digital cinema camera, 24mm wide-angle lens, f/5.6 aperture for deep depth of field. A fast shutter speed is used for the static elements, while the train in the mid-ground has a slight motion blur to convey speed. High dynamic range (HDR), 8k resolution. [Style & Texture] Modern urban aesthetic, J-pop or K-pop music video set style. Textures include gritty asphalt, weathered concrete buildings, smooth metallic train surfaces, and industrial matte yellow paint on the steel trusses. The color palette is dominated by "caution yellow," asphalt gray, and clear sky blue, creating a vibrant, high-energy atmosphere. [Negative Prompt] (Negative Prompt): human, people, girl, woman, dancers, face, body, movement, crowd, blurred figures, person in uniform, text, watermark, signature, messy, dark, indoor, night shot, low resolution, distorted perspective.