CREATEVISION AI · 实用视频指南
开始之前
01为每个独立场景使用同一参考；不要让身份漂移通过反复重新上传而累积。
02在更换场景时，保持面部结构、发型、身体和标志性配饰稳定。
03在将静止图像转为视频之前，先检查人物一致性。照片结果并不能证明视频一致性。
选定一张可重复使用的参考图
使用光线均匀、面部可见的肖像。避免重度滤镜、太阳镜和极端角度。我们的示例使用为演示而创建的虚构女性，并非真实人物的身份或代言。
在AI Influencer Generator中，每次上传同一张照片。该页面提供场景预设和服装/构图选项。如果使用单独的Characters库，请保存已确认的角色及其参考视图以供重复使用；那是一个独立的资产管理流程。
用同一人物制作五种场景
下面五个示例共享原始参考图。它们展示了咖啡馆、海滩、镜子自拍、产品开箱和酒店场景。服装可以不同，因此请将身份与服装分开评估。
观察下颌形状、眼睛、鼻子、发际线和身体比例。仅凭熟悉的发型是不够的。还要检查镜子中的手机、手指以及开箱场景中的产品。
阅读原始提示词
The subject is the exact person in the reference photo. Keep their face structure, eye shape and colour, nose, lips, skin tone, hairstyle and hair colour, body build and age recognisable in every detail — a follower who knows this person must recognise them instantly. Do not replace them with a generic model, do not change their age, do not slim or reshape their face or body, and do not add any other people to the scene. Scene: sitting at a window table in a bright specialty coffee shop, a latte with latte art and a half-eaten pastry on the wooden table, soft morning daylight from the window, relaxed half-smile at the camera as if a friend took the photo across the table. Outfit: casual everyday wear that suits the scene — a plain tee or knit, jeans or shorts, clean sneakers; no logos. Framing: vertical feed photo, head to hips. Shot like a real phone photo posted by a lifestyle creator: natural skin texture with pores and fine lines kept (no airbrushing, no plastic smoothing), realistic lighting with soft shadows, slight depth of field, candid framing. Photorealistic. No text, no captions, no watermark, no logos, no brand marks, no other faces in the frame.
阅读原始提示词
The subject is the exact person in the reference photo. Keep their face structure, eye shape and colour, nose, lips, skin tone, hairstyle and hair colour, body build and age recognisable in every detail — a follower who knows this person must recognise them instantly. Do not replace them with a generic model, do not change their age, do not slim or reshape their face or body, and do not add any other people to the scene. Scene: golden-hour on a quiet beach, sun low behind them, hair moving slightly in the breeze, sea and a soft horizon in the background, warm backlight with a gentle lens flare, carefree expression. Outfit: casual everyday wear that suits the scene — a plain tee or knit, jeans or shorts, clean sneakers; no logos. Framing: vertical feed photo, head to hips. Shot like a real phone photo posted by a lifestyle creator: natural skin texture with pores and fine lines kept (no airbrushing, no plastic smoothing), realistic lighting with soft shadows, slight depth of field, candid framing. Photorealistic. No text, no captions, no watermark, no logos, no brand marks, no other faces in the frame.
阅读原始提示词
The subject is the exact person in the reference photo. Keep their face structure, eye shape and colour, nose, lips, skin tone, hairstyle and hair colour, body build and age recognisable in every detail — a follower who knows this person must recognise them instantly. Do not replace them with a generic model, do not change their age, do not slim or reshape their face or body, and do not add any other people to the scene. Scene: full-length mirror selfie in a tidy bright bedroom or walk-in closet, holding a smartphone up at chest height with one hand, the phone partly covering the face is NOT allowed — face clearly visible, outfit fully shown, natural window light, casual confident pose. Outfit: relaxed streetwear — an oversized jacket or hoodie, wide trousers, chunky sneakers, a small bag; no logos. Framing: vertical feed photo, head to hips. Shot like a real phone photo posted by a lifestyle creator: natural skin texture with pores and fine lines kept (no airbrushing, no plastic smoothing), realistic lighting with soft shadows, slight depth of field, candid framing. Photorealistic. No text, no captions, no watermark, no logos, no brand marks, no other faces in the frame.
阅读原始提示词
The subject is the exact person in the reference photo. Keep their face structure, eye shape and colour, nose, lips, skin tone, hairstyle and hair colour, body build and age recognisable in every detail — a follower who knows this person must recognise them instantly. Do not replace them with a generic model, do not change their age, do not slim or reshape their face or body, and do not add any other people to the scene. Scene: holding up a small unbranded product (a skincare jar, a pair of headphones or a sleek gadget — no logos) toward the camera with an excited expression, sitting on a sofa or at a vanity with a ring-light glow in the eyes, a tidy bright living room behind. Outfit: casual everyday wear that suits the scene — a plain tee or knit, jeans or shorts, clean sneakers; no logos. Framing: vertical feed photo, head to hips. Shot like a real phone photo posted by a lifestyle creator: natural skin texture with pores and fine lines kept (no airbrushing, no plastic smoothing), realistic lighting with soft shadows, slight depth of field, candid framing. Photorealistic. No text, no captions, no watermark, no logos, no brand marks, no other faces in the frame.
阅读原始提示词
The subject is the exact person in the reference photo. Keep their face structure, eye shape and colour, nose, lips, skin tone, hairstyle and hair colour, body build and age recognisable in every detail — a follower who knows this person must recognise them instantly. Do not replace them with a generic model, do not change their age, do not slim or reshape their face or body, and do not add any other people to the scene. Scene: lounging by a luxury hotel rooftop pool with a city skyline behind, a drink on the side table, late-afternoon sun, polished editorial-travel feel, relaxed and unbothered. Outfit: polished evening look — an elegant dress or a fitted dark suit, subtle jewellery, well-groomed; tasteful, fully clothed. Framing: vertical feed photo, head to hips. Shot like a real phone photo posted by a lifestyle creator: natural skin texture with pores and fine lines kept (no airbrushing, no plastic smoothing), realistic lighting with soft shadows, slight depth of field, candid framing. Photorealistic. No text, no captions, no watermark, no logos, no brand marks, no other faces in the frame.
加载示例会填充草稿。生成前请检查设置和当前报价。
说明哪些必须保持不变
生产提示在场景描述之前先列出身份约束。它要求呈现自然的手机照片外观，而不是精致通用的模特。对于你自己的系列，仅在预设需要时添加一条简短的特定指令。
从已确认的原始图生成每个场景，而不是让每张新图像都依赖于一个逐渐漂移的先前结果。保留一张联系表，以便在整个系列中变化变得明显。
分别管理人物外观与穿搭
一致的身份可以穿不同的衣服，但产品宣传也可能需要相同的服装。在生成之前决定哪些细节是固定的。将参考图、所选场景、服装和构图与已确认的输出一起保存。
专用的influencer页面创建生活方式照片；单独的subject库可以存储可复用的参考图和角色表。不要假设每张照片输出都会自动成为已保存的角色，或转移到每个视频工具中。
把确认好的照片做成短视频
选择最强的一张静止图，打开image-to-video工作流。从一个小幅头部转动、克制的行走或轻柔的镜头运动开始。保持动作足够简单以便检查。
将已确认的图像带入视频草稿，并在运动过程中比较面部。如果使用角色替换模板，请检查所选场景是否支持你的角色的构图和身体比例。视频质量必须与照片系列分开审查。
为真正可用的系列做预算
先创建一个场景，在投入剩余四个之前先评估它。使用所选图像模型和质量的实时报价。更改模型或分辨率会改变成本，也可能改变外观。
这些示例记录了生产工作流和原始参考图。它们不是经过测量的身份成功率、自动发布系统或粉丝增长的承诺。请根据已确认的输出量选择方案，而不是你希望保留的生成次数。
生成前的问题
每个场景都能完美保留面部吗？
不能。参考约束有帮助，但你必须检查每一个结果。当面部漂移时，简化场景和造型变化。
我可以创建一个虚拟人物而不是使用自己的照片吗？
可以。先创建并确认一个虚构角色，然后在支持的工作流中复用其参考图。
照片预设会自动制作视频吗？
不会。influencer页面创建静止照片。使用image-to-video工作流来制作运动，并单独审查该结果。
比较方案和积分包
知道你需要多少用量？在选择之前，比较会员积分和一次性积分包。你的草稿始终由你掌控。
