正在打开每日关注
正在打开每日关注
作者分享在 Higgsfield 用 Seedance 2.5 生成 1:5 微缩人物便利店跑酷短片的完整结构化视频 Prompt。
Turning a convenience store run into an extreme adventure. Seedance 2.5 on @higgsfield Prompt : STYLE: Photorealistic live-action miniature-person FX, cinematic macro photography, ultra-shallow depth of field, realistic convenience store, dynamic handheld camera, dramatic practical lighting, realistic physics, detailed textures, premium viral commercial aesthetic, 4K. Create a 30-second cinematic adventure featuring one athletic miniature woman at approximately 1 : 5 scale. Maintain her exact face, hairstyle, outfit, body proportions, backpack, and scale throughout. She races across towering snack shelves, jumping over oversized packages and gaps. A giant candy box falls, creating an avalanche of rolling candies that she uses as stepping stones before landing on a massive chocolate bar. She climbs a huge drink bottle, runs up its curved surface, jumps over the cap, and lands on a hanging promotional display. She then rides a moving shopping cart through checkout, dodging the scanner and jumping onto the conveyor belt. At the refrigerated section, she runs across frozen-food packages, slides beneath a closing freezer door, and swings onto the display using the handle. Finally, she climbs onto a giant store sign and looks across the enormous store filled with products and shoppers as the camera pulls back to reveal just how tiny she is. Use dynamic macro tracking, dramatic low angles, smooth aerial-style movements, strong foreground parallax, realistic reflections, shadows, depth of field, product packaging, natural human movement, believable physics, seamless transitions, and cinematic lighting. End with the cinematic title: “SMALL SCALE. BIG ADVENTURE.” followed by “EVERY AISLE IS A NEW WORLD.” SFX: Footsteps, packaging crinkles, candy rattles, plastic flexing, cart wheels, scanner beeps, conveyor hum, freezer sounds, rushing air, and cinematic impact. NEGATIVE: No cartoon, anime, toy-like appearance, animation, duplicated character, face or outfit changes, distorted anatomy, floating, scale changes, narration, or subtitles.
是什么:一条可复用的 AI 视频生成配方。作者展示在 Higgsfield 平台用 Seedance 2.5 生成“微缩人物便利店大冒险”30 秒短片的完整 Prompt,并附两帧实拍感画面:微距浅景深下的货架攀爬与超市上空飞跃,验证了配方的实际效果。
核心设计:Prompt 采用清晰的结构化分层——STYLE 段锁定“实拍微缩特效 + 微距摄影 + 浅景深 + 手持运镜”的总体基调;剧情段按时间线描述连续动作(货架跑酷、糖果雪崩、攀爬饮料瓶、穿越收银台、冷柜滑行、结尾拉远揭示比例);再以镜头语言段(动态微距跟踪、低角度、前景视差)约束摄影表现;最后用 NEGATIVE 段排除卡通感、角色复制、比例漂移等 AI 视频常见翻车点。
关键洞察:视频生成已进入“导演式提示词”阶段——控制力的来源是把专业摄影术语与具体叙事动作逐一绑定,而非堆砌形容词;同时用负面提示词兜底一致性与物理可信度,这两点都可直接迁移到其他视频生成场景。
边界与风险:效果强依赖平台与模型能力,跨模型迁移性未知;30 秒长序列中维持角色面部与比例一致仍是难点;文中提到的 SFX 是否真实渲染取决于平台支持;商用需确认平台授权条款。