Gemini AI 图片提示词：可直接复制粘贴的实测模板

想找一条复制后就能用的 Gemini AI 图片提示词？先明确任务，再考虑风格。编辑上传照片时，关键是锁定人物或物体身份；从零生成新图，则需要把画面信息描述完整。这样能有效减少提示词过于笼统、成图千篇一律的问题。

Google 的官方 Nano Banana 提示词指南建议，先交代主体、场景、风格和整体氛围，再补充精确的编辑或文字要求。我还在 2026 年 8 月 11 日通过 AIReiter 的 Nano Banana Pro image model进行了三项小规模模型测试。下方结果仅作实用示例，并非通用基准；不同 Gemini 产品入口或账户，生成结果也可能不同。

30 秒选对 Gemini 图片提示词

如果必须保留真人、实物或老照片中原本的辨识特征，请先上传图片再编辑。若要从零创建虚构场景或通用主体，直接使用纯文字提示词即可，但提示词需要承担完整的视觉描述工作。

Google 的 Gemini Apps Help说明了图片生成与编辑功能。如果原始人脸被严重遮挡，或图片质量极低，建议先做小范围修改，再尝试完整转换；不同 Gemini 产品入口提供的模型名称和可用性也可能不一样。

已有照片可直接套用的提示词

以下 Gemini AI 图片提示词适用于你在同一聊天中上传的图片。稳定的写法很简单：先说明哪些内容绝不能变，再明确你只想改动哪一项。

换衣服和背景，但人物不能变

复制下面的提示词后，替换方括号中的内容。无论想换什么服装或地点，身份锁定部分都建议保留。

Use my uploaded photo as the identity reference. Keep my exact facial features, age, skin tone, hairline, hairstyle, body proportions, and natural expression unchanged. Change only my outfit to [describe the outfit] and the background to [describe the location]. Keep the same person, pose, camera angle, and approximate body position. Match the new lighting to the background so the edit looks like one natural photograph. Do not beautify, slim, de-age, add accessories, or change my identity. Photorealistic result, [aspect ratio].

如果第一张结果改了你的脸，不必重新粘贴整段提示词，直接用下面这条针对性补救：

Edit the current result only. Restore the facial geometry, age, skin tone, hairline, and expression from my uploaded reference photo. Keep the outfit and background exactly as they are. Change nothing else.

修复并上色老照片

老照片修复不是“美化改造”。要明确告诉 Gemini：它的任务是补回受损信息，同时保留原始表情、脸型、服装和画面构图。

Restore the uploaded old photograph. Remove scratches, tears, dust, stains, fading, and small film damage. Reconstruct missing edges conservatively and preserve the original facial geometry, expression, age, hairstyle, clothing, pose, camera angle, and composition. Colorize it with restrained, period-appropriate colors and natural skin tones. Do not modernize the clothing, smooth the face, change the background, add details that are not supported by the photo, or make the person look younger. Keep the result photographic and faithful to the source.

对于模糊原图，建议额外加上这句，而不是要求模型凭空造出一张“完美”人脸：

Improve clarity only where the source contains recoverable detail; do not invent a new identity or sharpen facial features beyond what the reference supports.

用参考照片生成情侣或家庭场景

涉及多个人物时，应为每张参考图编号，并写明最终画面中应有几个人。这样更容易发现并修正换脸、多出人物等错误。

Use the uploaded images as identity references. Place the person from photo 1 and the person from photo 2 together in [describe the setting], naturally [describe the action]. Preserve each person's exact facial features, age, skin tone, hair, body proportions, and recognizable expression. Keep the clothing from photo 1 and photo 2 unless I explicitly request a change. Include exactly two people. Do not add, remove, merge, or swap identities. Use [lighting and photographic style], [camera framing], and [aspect ratio].

如果是三人或更多人，将“exactly two people”替换为正确人数，并依次标记 photo 1、photo 2、photo 3。再次修改风格前，先检查脸部和手部是否正常。

毕业照或生日人像

想做一张可信的庆祝照片，而不是套上泛泛的风格滤镜，可以用这条简洁的场景模板。

Use my uploaded photo as the identity reference and place me in a natural [graduation / birthday] portrait. Keep my exact facial features, age, skin tone, hair, body proportions, and expression unchanged. Add [cap and gown / birthday outfit], [simple celebration setting], and [warm natural lighting]. Keep one person, a believable pose, and realistic hands. Do not beautify, de-age, add extra people, or change my identity. Use a [4:5 / 3:2] composition, no text, no watermark.

从零生成 Gemini 图片的可复制提示词

纯文字提示词没有参考照片来固定画面，因此需要补全主体、动作、场景、构图、光线和输出格式。

写实职业头像

可替换人物、服装和背景描述，同时保留光线与构图限制。这是一条模型级测试提示词，不是 Gemini App 的截图。

Create a realistic professional headshot: a woman in her early 30s with natural skin texture and shoulder-length dark hair, wearing a navy blazer over a white shirt, facing slightly toward a large window in a quiet modern office. Use soft window light from camera left, a neutral background, an 85mm portrait-lens look, and a shallow depth of field. Keep realistic proportions and a clean 4:5 composition. No text, no watermark.

AIReiter nano_banana_pro 在 4:5 下的测试：该样张呈现出自然的窗边光线，主体与背景分离清晰。

带留白的电影感旅行人像

留白只有和构图、人物位置绑定时才真正有用。说明人物处于画面哪个位置，以及空白区域准备留作什么用途。

Create a photorealistic travel portrait: a person in a rust-colored jacket standing on a quiet coastal cliff at blue hour, with wind moving the fabric slightly and distant sea haze behind them. Use a restrained teal-and-amber color grade, documentary travel photography, a 35mm lens perspective, and natural skin texture. Place the subject on the right third with open negative space on the left. Use a 3:2 landscape composition. No text, no watermark.

AIReiter nano_banana_pro 在 3:2 下的测试：人物保持在右侧三分线位置，左侧保留了可用于加标题或裁切的空间。

标签清晰可读的产品图

精确文字要放进引号，并说明它在画面中的位置。文字尽量简短：即便如此，生成图片中的小尺寸标签仍可能出现拼写错误或变形。

Create a clean photorealistic product hero image for a matte black reusable water bottle on a pale stone table, with one green leaf beside it and soft morning window light. Add a believable contact shadow, a minimal warm-grey background, premium studio product photography, a front three-quarter view, and a 4:5 portrait composition. Add the exact readable label text "FIELD / 01" on the bottle, centered and spelled exactly. No other text, no watermark.

AIReiter nano_banana_pro 在 4:5 下的测试：这张样图中的指定标签清晰可读，但不代表文字渲染一定完全准确。

如果标签文字不对，粘贴以下追问：

Edit only the label. It must read exactly "FIELD / 01" on two lines, with FIELD on line 1 and / 01 on line 2. Keep the bottle shape, lighting, leaf, table, framing, and every other pixel unchanged. Check the spelling before returning the image.

杂志感胶片人像

应先写清主体和构图，再加入风格参照。这样“胶片感”只会改变画面处理，而不会替换人物或场景本身。

Create a realistic editorial portrait of [subject] in [specific location], standing [describe pose or action]. Frame the subject as a [close-up / waist-up / full-body] portrait with [camera angle] and [aspect ratio]. Use [light source and direction], natural skin texture, and believable fabric detail. Give it a restrained 1990s film-photography look with subtle grain, gentle highlight roll-off, and slightly faded colors. Keep the scene photographic, not illustrated; no text, no watermark.

Y2K 数码相机抓拍

如果你想要的是有辨识度的 2000 年代初期随拍质感，而不是精修棚拍人像，可以使用这条提示词。

Create a realistic early-2000s compact-digital-camera snapshot of [subject] at [specific place], wearing [outfit] and [natural pose]. Use direct on-camera flash, slight motion softness, modest sensor noise, imperfect framing, cool night colors, and a small amount of authentic red-eye. Keep the subject's face and body proportions natural. Make it feel like a real candid photo, not a polished fashion image. Use a 4:5 portrait composition, no watermark.

微缩模型或手办照片

与其只写“miniature”，不如加入明确的比例参照，这样微缩效果会更容易被模型正确理解。

Turn [the subject or uploaded photo] into a highly detailed collectible figurine displayed inside a backlit glass cabinet. Make it clearly miniature, about 15 cm tall, standing on an engraved base with a human hand reaching in to adjust it for scale. Keep the subject's recognizable facial features and clothing details. Show softly blurred figurines on shelves behind it, warm display lighting, shallow depth of field, and a photorealistic finish. No text except the small engraved base, no watermark.

Gemini 图片常见翻车问题：修复提示词

当第一张结果只错了一项要求时，再追加一条修复提示词即可。修改范围越窄，越方便逐项确认问题是否解决。

脸已经不像参考图了

Use the uploaded reference as the source of truth for identity. Restore the person's facial geometry, age, skin tone, hairline, hairstyle, and natural expression. Keep the current pose, clothing, background, lighting, and composition unchanged. Change only the face so it matches the reference; do not beautify, slim, de-age, or reinterpret the person.

画面太平，像卡通

Refine the current image without changing the subject or composition. Replace the flat, illustrated look with physically believable light, natural material texture, realistic lens perspective, soft but directional shadows, subtle depth of field, and small photographic imperfections. Keep the person's identity, clothing, pose, and background unchanged. Do not add a glossy filter or extra detail that is not supported by the scene.

TechRadar 曾引用一位 Reddit 用户对这一问题的反馈：

"Now, almost every image looks flat or cartoonish, no matter how detailed my prompt is." - TechRadar, 2026 年 8 月 5 日

这只是用户反馈，并非基准测试；真正可操作的部分是上面的修复提示词。

Gemini 多加了人物，或改了衣服

Edit the current result only. Keep exactly [number] people, with the same identities, faces, hair, body proportions, and clothing as the reference. Remove any extra person or duplicate limb. Do not change the background, pose, lighting, or camera framing. Apply only this requested correction: [one correction].

图片里的文字不对

Edit only the visible text. The text must read exactly "[YOUR TEXT]". Keep the same number of lines, spelling, punctuation, font style, position, perspective, and contrast needed for legibility. Do not change the object, background, lighting, or composition. Perform a final spelling check before returning the image.

这些模板背后的可编辑提示词公式

一条可复用的 Gemini AI 图片提示词，本质上可以拆成六项：主体、动作、场景、构图、光线/风格和限制条件。

Google 官方指南的建议也是如此：描述人物、场景、风格与氛围；涉及编辑、多张图片或精确文字时，要求必须写得明确。

先锁定不变项，再提出改动

处理上传照片时，可在提示词开头附近使用这个句式：

Keep [identity, object, or layout facts] unchanged. Change only [the requested field].

单独写“Keep my exact facial features”不如完整约束有效，例如：“keep my facial geometry, age, skin tone, hairline, hairstyle, body proportions, and expression unchanged.” 下一轮修改时，每次只改一个字段，避免已经成功的部分被意外重新生成。

Gemini 图片提示词 FAQ

使用 Gemini AI 图片提示词一定要上传照片吗？

不一定。纯文字提示词可以生成新的人物、产品或场景。只有在需要 Gemini 编辑或保留特定人物、物体或老照片时，才需要上传图片。

怎样让 Gemini 不要改我的脸？

使用换装提示词中的身份锁定语句：脸部几何特征、年龄、肤色、发际线、发型、身体比例和表情。如果第一次结果出现跑偏，可以要求只修正脸部，同时锁定当前服装、场景和光线。

这些提示词能用于 Nano Banana 吗？

可以，前提是你所使用的 Gemini 产品入口中，Nano Banana 是可用的图片生成或编辑模型。Google DeepMind 的提示词指南使用 Nano Banana 这一名称，而 Gemini Help 则说明了当前 Gemini Apps 的图片工作流程；不同产品入口中的模型名称和可用性可能会变化。

为什么 Gemini 返回的图片几乎没有变化？

可能是你要求的改动太细微，也可能是提示词里的保留要求多于变更要求。明确写出一个可见变化、它应该出现在什么位置，以及哪些内容必须保留；随后在追问中只要求执行这一项修改。

应该上传几张参考照片？

使用能够证明这次编辑需求的最少图片即可。单人转换时，一张清晰参考图通常更容易检查；情侣或家庭合成时，请为每张图片编号，并明确写出最终结果需要包含的准确人数。