一条文字提示词,一次生成:画面、人声和街头环境声都在同一个文件里,每秒 $0.10。想要一镜到底就写明,否则 Omni 会自己把短片切成几个机位;它也认 [0-3s] 这样的时间码和“Audio/sound:”这一行。
模型
Gemini Omni 1.1 Flash
画幅
9:16 · 10 s · 720p
本次生成
$1.00
生成用时
2–5 分钟
实际使用的提示词
Gemini Omni 1.1 Flash · 9:16 · 10 s · 720p · $1.00
Vertical street interview, handheld phone footage. A friendly interviewer's hand holds a small microphone toward a woman in her thirties on a sunny city sidewalk. She holds up a plain white tube of hand cream with no label and says, smiling: "Honestly? I bought it because of the smell, and now I carry it everywhere." Natural street ambience, light traffic, shallow depth of field.
写清楚发生了什么,把台词引出来,镜头就有了。这段 Wan 3.0 的戏是从一张画面开始的。需要更长的篇幅时,Wan 一次生成最长 30 秒,声音不额外收费,还支持 480p 草稿和用视频输入接着往下拍。
模型
Wan 3.0
画幅
9:16 · 10 s · 720p
本次生成
$1.00
生成用时
2–5 分钟
实际使用的提示词
Wan 3.0 · 9:16 · 10 s · 720p · $1.00
Locked-off vertical close-up. The woman in the magenta wool coat stands still in the dim windowless apartment hallway, the warm wall lamp is already on and stays at constant brightness. She holds the small folded paper note in both hands at chest height, reads it for two seconds, slowly lowers her hands to her waist, looks up past the camera, startled, and whispers clearly: "He knew." She holds the look, breathing quietly. Nothing in the background moves or changes: the lamp, the plain wall and the closed door stay exactly as they are. Quiet room tone, faint distant thunder. No text visible on the note, no other people, no camera movement.
这三张画面是同一个人物:一张参考人像、由它生成的新场景,以及以该场景为首帧的说话短片——Grok Imagine Video 1.5,720p。每次生成最多 14 张参考图,每条提示词最多 4 张。仅限 SFW 的信息流内容。
实际使用的提示词
Nano Banana Pro · 4:5 · 2K · $0.11
The same woman as in the reference image, identical face, green eyes and long wavy honey-blonde hair now tied in a high ponytail, an adult of 24, athletic hourglass figure: fitness influencer photo in a bright modern gym. She wears a matching dusty-rose sports set — a supportive sports top and high-waisted leggings made of smooth matte solid-colour fabric with no ribs, no stripes, no mesh and no knit texture — standing relaxed next to a rack of dumbbells, a plain smooth white towel over one shoulder, friendly confident smile at the camera, morning light from large windows. Three-quarter framing, natural phone-camera look, tasteful, no text.
Gemini Omni 1.1 Flash · 9:16 · 10 s · 720p · 有声 · $1.00
Vertical street interview, handheld phone footage. A friendly interviewer's hand holds a small microphone toward a woman in her thirties on a sunny city sidewalk. She holds up a plain white tube of hand cream with no label and says, smiling: "Honestly? I bought it because of the smell, and now I carry it everywhere." Natural street ambience, light traffic, shallow depth of field.
街头采访之二:换个人,换个产品
Gemini Omni 1.1 Flash 9:16 · 10 s · 720p · 有声
$1.00
实际使用的提示词
Gemini Omni 1.1 Flash · 9:16 · 10 s · 720p · 有声 · $1.00
Vertical street interview, handheld phone footage. A friendly interviewer's hand holds a small microphone toward a man in his thirties with a short beard on a busy pedestrian street on an overcast afternoon. He holds up a plain matte black reusable water bottle with no logo and says, laughing: "My wife bought it for me as a joke. Now I refill it four times a day." Natural street ambience, passing footsteps, shallow depth of field.
自拍测评:产品贴着脸,一句清楚的台词
Wan 3.0 9:16 · 8 s · 720p · 有声
$0.80
实际使用的提示词
Wan 3.0 · 9:16 · 8 s · 720p · 有声 · $0.80
Vertical selfie-style UGC review, handheld phone front-camera look, starting exactly from the first frame. The woman in the bright bathroom keeps holding the plain white pump bottle with no label next to her face with one hand, keeps it still and upright, and says to the camera clearly, with a warm smile: "I was skeptical, but two pumps in the morning and my skin stays soft all day." She says this one sentence only once, finishing by the sixth second, then stops talking, closes her lips and keeps smiling at the camera in the same pose as the first frame until the end of the shot. Her appearance, age, hair and grey t-shirt stay exactly as in the first frame. Only slight natural handheld sway, framing unchanged, the bottle keeps the same shape, her fingers stay natural. Natural room tone, soft daylight, shallow depth of field. No music. No text on screen.
桌面开箱,带口头反应
Grok Imagine Video 1.5 9:16 · 6 s · 720p · 有声
$0.84
实际使用的提示词
Grok Imagine Video 1.5 · 9:16 · 6 s · 720p · 有声 · $0.84
Vertical UGC unboxing, phone camera on a small desk tripod, slightly high angle. A man in his twenties sits at a wooden desk, lifts the lid off a plain kraft cardboard box, takes out a matte black wireless earbuds case with no logo, holds it up to the camera and says happily: "Okay, this feels way more expensive than it was." Natural room sound, soft cardboard rustle. Both hands move naturally. Plain wall behind him, nothing in the background moves. No text, no logos.
图片需 1–2 分钟;视频需 2–5 分钟,取决于模型和设置。Veo 3.1 Fast 和 Veo 3.1 Lite 通常更快。生成期间可以关闭页面稍后回来——结果会在你的账户里等着。
图像生成
6 个回答
BananaBanana 的角色一致性是怎么工作的?
BananaBanana 内置 角色一致性功能——上传一张人像照片作为角色参考,描述你的场景,AI 会在把这个人置于新场景、新服装或新艺术风格的同时保留面部特征。每次生成最多可附加 14 张参考照片(角色和物品),在多张图片间保持角色一致。适合 AI 网红内容、社交媒体视觉、营销素材,或任何需要同一张脸出现在不同场景的项目。
可以用 BananaBanana 制作 AI 网红或社交媒体内容吗?
可以——BananaBanana 是按 AI 网红和内容创作者的工作流搭建的,提供参考模式、角色一致性和每条提示词最多 4 张图片。做一个稳定的 AI 角色,再让同一个角色出现在不同场景、服装和地点——适合 Instagram、TikTok 等平台。Nano Banana Pro 能出非常逼真的专业级人像和生活方式照片。与 Google Gemini 网页版不同,BananaBanana 生成的内容没有可见水印。注意:Google 模型(Nano Banana、Veo、Gemini Omni)生成的图片和视频带有 Google 不可见的 SynthID 水印——它不影响画质,但可能被某些平台的 AI 内容识别系统检测到。开启社交媒体优化,可以在不损失画质的前提下,把图片处理成更适合在社交平台干净发布的样子。按量付费、用加密货币或银行卡充值,个人创作者和机构都负担得起。