FLOW AI CONSULTING· 客户教学
正在爆流量的
AI 视频制作
学会这个就够了!
AI 视频破屏效果成品示例

GPT Image 2.0 + Seedance 2.0
爆款 AI 视频制作指南 在 Higgsfield 上:一张图 → 一段会「破屏而出」的电影感视频

核心思路:先用 GPT Image 2.0 出一张高质量图,再用 Seedance 2.0 让它动起来。 图是地基,地基越稳,视频越好。

使用工具:Higgsfield(同一平台内切换) · 图像模型 GPT Image 2.0 · 视频模型 Seedance 2.0
STEP 1用 GPT Image 2.0 出基底图
  1. 进入 Higgsfield,顶部切到 图像(Image)
  2. 点模型按钮(默认 GPT Image 2.0)→ 打开模型下拉
  3. GPT Image 2.0
  4. 用输入框左侧 + 上传一张基底照片(可选,但建议)
  5. 粘贴下方图像提示词 → 点生成,等 30 秒~1 分钟
建议设置:比例 9:16 · 质量 4K(图越清晰,后面视频越清晰)
图像提示词 · GPT IMAGE 2.0复制
A hyper-realistic first-person POV photograph of a dimly lit desk with a dark mechanical keyboard and a round speaker. The user's hand hovers 2 centimeters away from the monitor glass. The monitor screen is very bright, displaying an ethereal, goddess-like pale woman with long silver-white hair. She radiates a divine, mystical aura. She wears a delicate thorn crown and a solid black blindfold. Her neck and torso are elegantly wrapped in dark, woven thorny vines forming a minimal bodice against her pale skin. Behind her left shoulder is a majestic, large dark wing. The background inside the screen is a bright, luminous, sunlit snowy rocky landscape with subtle warm yellow moss on the stones, giving a bright daytime feel. The intensely bright, holy light from the screen illuminates the dark room and the hovering hand. Windows taskbar is visible at the bottom of the screen. Cinematic lighting, photorealistic
自己的桌面实拍基底照示例

▲ 用自己的桌面实拍照做基底,代入感更强(图为 Flow Ai 落地页实拍)

📸 基底照怎么拍

「在光线昏暗的房间里,坐在桌前、手伸向显示器」—— 用手机第一人称视角拍。

不上传基底照、只输入提示词,在 Higgsfield 上也能出类似效果;但用你自己的桌面照,出来的代入感会强很多。

STEP 2切到视频,用 Seedance 2.0
  1. 顶部切到 视频(Video)
  2. 用输入框左侧 + 上传刚生成的图(或在图库点图 → 选「生成视频」)
  3. 点模型按钮(默认 Seedance 2.0)→ 选 Seedance 2.0
  4. 粘贴下方视频提示词 → 点生成,等 2~5 分钟
建议设置:比例 9:16 · 时长 5 秒或 10 秒 · 分辨率 720p(1080p 更耗 credit)
侧边模型可选:Pollo AI / Google / Sora / Wan AI / Kling AI / Happy Horse / Hailuo AI 等。
视频提示词 · SEEDANCE 2.0复制
Starting from the provided image. The user's hovering hand moves forward to touch the monitor glass. The ethereal woman on the screen reacts, raising her hand to mirror the user's movement. Their fingertips meet at the glass surface. Upon contact, the screen ripples like liquid, emitting a radiant, holy white glow. Suddenly, her black fabric blindfold cracks and falls away, revealing glowing pale eyes. Her expression shifts to awe. She tightly grabs the user's hand. The user pulls back sharply. As the user pulls, her arm completely breaches the screen boundary, the monitor glass shattering outward into sharp jagged shards mixed with crystalline ice fragments and a puff of pale frost mist, transitioning seamlessly from a flat 2D image into a hyper-realistic 3D volumetric form inside the dark room. Her pale skin, the thorny vines on her body, and her dark wing gain realistic depth, texture, and shadows. Shimmering glass-like digital debris, icy shards, and glowing particles float around the breach. She leans forward, her shoulders emerging from the intensely bright monitor into the dimly lit room. The first-person POV camera shakes dynamically as the hand is pulled back. Cinematic shallow depth of field, rapid focus shift from the screen to the emerging 3D figure. High contrast between the bright divine light of the screen and the dark room.

Audio: a sharp glass-shattering crack with crystalline ice tinkling at the moment the arm breaches the screen, layered with a deep sub-bass impact.
STEP 3查看结果 + 重生成

视频不满意就点「重新生成」再跑一次 —— Seedance 同一条提示词每次结果都不一样,多跑几次挑最好的。

🎭 想换角色风格?只改这几处

不用重写整段,只替换图像提示词里这几个词,就能换成完全不同的世界观:

原版变体示例
goddess-like pale woman with silver-white hairdark elf warrior with glowing red eyes
暗黑精灵
thorn crown and black blindfoldgolden ornate crown with floating runes
符文王冠
dark woven thorny vinesglowing electric circuits
赛博朋克
sunlit snowy rocky landscapevolcanic lava cavern with floating embers
熔岩洞窟
large dark wingmechanical robotic wings with neon light
机械翅膀

💬 结果不满意时,怎么跟 AI 沟通

核心原则:别只说「不好」,要说「哪里、怎么不对」。图像/视频生成 AI 不是聊天型,而是「改提示词」型 —— 要自己分析问题出在哪,再去改提示词。

不同情况,往提示词里加这些词:

问题加进提示词 / 修改
角色太平淡intricate details, detailed costume, ornate jewelry, fine fabric textures
画面太暗bright neon glow from screen, strong rim light
碎屏效果太弱强调 explosive shatter, hundreds of sharp shards flying outward
手很奇怪realistic anatomically correct hand, natural finger positioning
第一人称视角不对强调 strict first-person POV from user's eye level, no face visible
角色只待在屏幕里视频提示词里加 she steps fully out of the screen

🛠️ 常见卡点 & 解决法

Q1. 角色的手很奇怪(六根手指等)
简单版(直接说)
手的形状很奇怪,请把手重新做得自然一点。
详细版(往原提示词里加)
ADD复制
perfectly proportioned human hand with exactly five fingers, anatomically accurate, no extra digits, clean natural pose
Q2. 屏幕里画面模糊 / 看不清
简单版
希望屏幕里的角色更清晰一点。
详细版(往原提示词里加)
ADD复制
ultra-sharp focus on monitor screen content, high contrast display, vivid bright colors inside the screen
Q3. 视频太静(角色不动)
简单版
角色太安静了,让它动得更有张力一点。
详细版(加在视频提示词最前面)
ADD(开头)复制
Highly dynamic motion throughout the entire clip. Constant camera movement and character action.
Q4. 碎屏效果太弱
简单版
把屏幕玻璃破碎的场面做得更华丽。
详细版(往视频提示词里加)
ADD复制
explosive glass shatter with hundreds of sharp shards, slow motion debris, sparkling reflective fragments, dramatic particle effects, cinematic VFX quality

💰 省 Credit 技巧

  1. 先把图打磨好再做视频 —— 视频比图耗 credit 约 10 倍。在图像阶段反复重生成 5~10 次,挑到最满意的那张,再去做视频。
  2. 先 720p,满意再 1080p 重生成 —— 分辨率直接影响 credit 消耗。
  3. 从短视频开始 —— 先用 5 秒确认概念,满意再扩成 10 秒。
  4. 每天领免费 credit —— 每天登录会发放一定额度的免费 credit。

爆款不是靠运气,
是靠一套能复制的流程

把流程跑顺、做成可复用的模板 —— 这是 Flow AI Consulting 想带你做到的。

已复制 ✓