你好 👋
这套流程能让任何一张照片「动起来」开口说话,从静态图到一条可发布的竖版短视频,全程大约 30 秒。
下面我把每一步、以及最关键的那段提示词,完整拆给你 —— 照着做就能复现。
源图的清晰度,直接决定最终成片的清晰度。所以第一步永远是「先把图做到极致」。
同一套流程,靠不同的 Prompt 还能做到:
✅ 更换背景 ✅ 更换场景 ✅ 添加特效 …… 玩法几乎无限。
这是整套流程的灵魂。它的作用是 锁死参考图的一切(人物、环境、构图),只让动作跟随驱动视频 —— 避免 AI 自己「乱改」。
场景:你已有一张 AI 成图,想让它跟着驱动视频开口、运动。
Match the reference image exactly. Preserve 100%: identity, face, eyes, nose, mouth, hairstyle, skin tone, body shape, clothing, accessories, chair, laptop, desk, wall, background, lighting, camera angle, composition, framing. Keep every object exactly in the same position. Do not change the pink chair. Do not change the laptop. Do not change the desk. Do not change the wall. Do not change the room. Follow only the motion from the driving video. Animate the character only. The environment must remain completely identical to the reference image. No background replacement. No background generation. No background reconstruction. No scene change. No object change. No object removal. No object addition. Maintain perfect facial consistency. Ultra realistic. Natural speaking motion. Natural facial expression.
提示:上面列举的物体(pink chair / laptop / desk…)可按你参考图里的实际物体替换,原理是把场景里每个元素都「点名锁定」。
场景:你已有一段现成视频,想把出镜人物替换成你本人 / 你的 AI 形象,并做到真实无破绽。
Replace the person in the reference video with the woman from the uploaded image. Keep her identity, facial features and hairstyle. Match the same camera angle, pose, framing and lighting as the target image. Photorealistic photo. Natural skin texture. Natural hair texture. Remove microphone. Negative: cutout, sticker, pasted, outline, halo, mismatched lighting, floating person, collage, extra people, anime, cartoon, illustration, watermark, text
说明:「uploaded image」请用你本人或你自己的 AI 形象。Negative 部分是为了避免「贴图感、抠图边缘、光线不搭」等常见破绽,让换入的人物自然融入画面。
背景、特效也可以靠提示词进一步调整 —— 同一套方法,能延伸出无数种风格。
AI 不会取代你。
但「会用 AI 的人」会取代你。🔥
现在就开始动手 —— 这正是 Flow AI Consulting 想带你做到的事。