由 Google 出品

Veo 3.1 提示詞

Google 的电影级文生视频模型——像导演一样指挥镜头,声音与画面一次生成。

video
First Frame Image Prompt A realistic, eye-level street view of a busy New York City intersection during the day. In the background, a massive billboard covers the side of a tall, classic brown brick building with visible windows. The billboard features a high-fashion, hyper-realistic image of a handsome man with short, tousled brown hair lying on his side against a backdrop of a light blue sky with fluffy white clouds. He is wearing a white button-down shirt with blue detailing and dark denim jeans. To his right on the billboard is a dark amber glass jar with a black lid and a white label. In the foreground, a bright yellow NYC taxi is driving past, and several pedestrians in casual clothing are walking across a crosswalk. There is a green street sign and scaffolding below the billboard. Natural sunlight, photorealistic style. Image-to-Video Prompt Theme: Faux out-of-home (FOOH) CGI advertisement where a billboard model comes to life and throws a product into the real world Visuals: Busy NYC street corner, massive billboard on a brick building showing a fashion model lying against a cloudy sky backdrop with a scented candle jar beside him; yellow taxi, pedestrians, green street sign, scaffolding, photorealistic urban daylight Camera: Static POV from a pedestrian on the street, slight rack focus shift at the end toward the caught candle in the foreground Style: Photorealistic, cinematic, hyper-realistic CGI blending 2D billboard illusion with 3D pop-out effect Action + Sound Design: The man on the billboard remains still for a brief moment as street traffic moves naturally in the foreground — ambient city sounds, car engines, distant chatter He seamlessly transitions from a static 2D image to a moving 3D figure, sitting up smoothly and crossing his legs left over right, revealing polished black leather boots, while smiling confidently toward the camera He reaches down with his left hand, picks up the dark amber candle jar beside him, and holds it up as if preparing to offer it He tosses the candle forward, out of the billboard’s frame, creating a 3D pop-out illusion as the jar flies toward the viewer A real human hand reaches up from the bottom of the frame and catches the candle; the camera pulls focus to the jar in the foreground, revealing the white label reading “SCENTED CANDLE,” “PRESS,” “No 5,” and “KEY NOTES: Ginger, Verveine, Vanilla,” while the billboard blurs softly in the background Ambient street traffic continues naturally throughout the entire sequence.
{ "duration": "10s", "aspect_ratio": "16:9", "fps": 30, "style": "photorealistic premium tech commercial, cinematic macro photography", "scene": "A 3-inch-tall adult male presenter in black streetwear with subtle neon-cyan accents interacts with gigantic premium black wireless headphones on a white marble studio desk.", "action": [ "0-3s: The miniature presenter confidently runs along the curved matte-black aluminum headband. Ultra-low macro tracking camera follows directly behind him.", "3-6s: He reaches an oversized knurled metal volume dial, grips it with both hands and forcefully rotates it. A subtle translucent sonic shockwave expands from the headphones.", "6-8s: He jumps from the headband onto the giant plush black leather memory-foam ear cushion, which visibly compresses on impact.", "8-10s: He stands on the cushion, smiles at the camera and gives a quick thumbs-up as the camera rapidly cranes upward and pulls back, revealing the full headphones, laptop and coffee cup on the desk." ], "camera": "Extreme macro shallow depth of field, smooth tracking, controlled whip-pan to the dial, dynamic follow-through on the jump, then rapid crane-up pullback into a wide hero shot.", "lighting": "Soft studio lighting with subtle cyan and electric-magenta rim lights, realistic metal, leather and marble reflections.", "audio": { "music": "NONE", "dialogue": "NONE", "voiceover": "NONE", "effects": "Realistic sneaker squeaks, mechanical dial clicks, deep rotation sound, subtle sonic whoosh and soft leather cushion impact." }, "quality": "8K-quality detail, realistic human anatomy, physically accurate materials and physics, stable character and product design.", "negative": "no music, no dialogue, no text, no logos, no watermark, no extra people, no floating, no teleportation, no clipping, no distorted hands, no changing scale, no character or clothing changes, no cartoon CGI, no excessive motion blur, no camera shake." }

關於 Veo 3.1

Veo 3.1 是 Google DeepMind 的旗舰视频生成模型,能把一段文字场景描述变成电影感短片。与多数视频模型不同,它原生生成音频——对白、环境声、音效与画面在同一次生成中同步完成。

它最突出的能力是导演级控制。Veo 理解电影语言:景别、镜头选择、推轨与升降等运镜、布光方案与情绪氛围。写得像分镜脚本的提示词,效果稳定优于松散的形容词堆砌。

正因如此,提示词质量在 Veo 上比在多数模型上更重要。下方画廊收录了经过验证的 Veo 提示词及其真实输出,你可以直接复制一个可用的结构,而不必从零摸索。

如何寫好 Veo 3.1 提示詞

  1. 1按分镜简报的结构写:主体 → 动作 → 场景 → 运镜 → 光线 → 情绪。Veo 偏爱电影词汇而非形容词堆叠。
  2. 2明确写出运镜——「缓慢推轨」「手持跟拍」「航拍降落」。每个片段只给一条运镜指令,动态才连贯。
  3. 3同时指挥声音:对白放引号里,再描述环境声(「雨打铁皮屋顶,远处闷雷」)。不写音频提示,得到的往往是平庸的环境音。
  4. 4用光线和调色参考锚定画面质感:「黄金时刻逆光,浅景深,电影感青橙调色」。
  5. 5一条提示词只写一个场景。Veo 擅长单一连续镜头,把多个剪辑塞进一条提示词会让运动碎裂。
  6. 6每次只改一个变量——运镜、光线或动作——你才能判断到底是哪句话改变了结果。

常見問題

Veo 3.1 最擅长什么?

带真实运动感和同步原生音频的电影级单镜头短片。当你需要「被导演过」的画面——有意图的运镜、连贯的光线、与场景匹配的声音——它是最强选择。

怎样写好 Veo 3.1 提示词?

像导演一样写,而不是像打标签。用自然的句子描述一个场景,明确写出运镜、光线和声音提示。本页经过验证的提示词是比从零开始更快的起点——复制一条,换成你的主体即可。

Veo 3.1 能生成声音和对白吗?

能——原生音频正是 Veo 的招牌能力。它生成与画面同步的语音、环境声和音效。把台词放进引号,并明确描述声场,效果最佳。

在哪里可以试用这些 Veo 提示词?

Veo 可在 Google 的 Gemini 应用与 Flow 中使用,开发者可走 Gemini API。从本页复制任意提示词粘贴过去,替换主体或运镜描述即可适配你的场景。