我搜集了全网最强的 MiniMax H3 提示词库,原来高手的视频是这么写出来的
我搜集了一个全网最强的 MiniMax H3 提示词库。
550 条提示词,每一条都带成片,每一条都能回到原作者的帖子。 看片、复制、粘贴,直接跑。
先看库里的一条:
公园巨型锦鲤(原作者成片)
西装大叔蹲在池塘边喂锦鲤,一条巨型锦鲤窜出水面,咬住面包袋,把他整个人拖进了池塘。手抖、跑焦、水花,像极了路人随手拍的。
H3 大家都跑过了。但把这 550 条翻一遍你会发现:出片好的人,提示词早就不是一段画面描述,而是一份分镜脚本。 几秒发生什么、镜头怎么动、谁的衣服不许变,连"不许出什么错"都提前写好了。
同样用 MiniMax H3,有人做出广告片,有人做出 PPT——差距不在模型,在提示词。
这篇把库里最值得抄的写法拆给你看:一条好提示词的 6 个结构,广告片的三段式,Vlog 的"设定 + 镜头清单",以及参考素材的分工写法。最后再说怎么把它们交给 API 和 Agent 批量跑起来。
H3 的提示词,其实是一份分镜脚本
先回顾 H3 官方文档里的几个关键参数:
- 输出 768P 或 2K,单条 4 到 15 秒;
- 参考素材最多 9 张图、3 段视频、3 段音频,合计不超过 12 个文件;
- 提示词上限 7000 个字符。
7000 个字符是什么概念?差不多是一页半的分镜脚本。
这意味着 H3 不只是“描述一下画面”的模型。你可以告诉它:谁在拍、几秒发生什么、镜头怎么动、哪些东西不许变、声音是什么、哪些翻车要避开。
你写得越像导演,它拍得越像片子。
我们统计了库里的 550 条提示词,长度的中位数大约 2000 个字符。能出好片的提示词,普遍不短。
但问题也在这里:大多数人不知道该写什么,写长了也只是在堆形容词。最快的学法,是直接看别人写好、而且已经跑出成片的提示词。
这个库能帮你省掉什么

省掉试错:每一条都先看片,再决定抄不抄。 550 条提示词,每条都配了生成出来的视频。效果好不好,点开就知道,不用自己花钱一条条跑。
省掉找原帖:每一条都能回到源头。 550 条都回查过 X 原帖:原帖明确提到 MiniMax H3,并附了生成视频。其中 341 条的完整提示词就在原帖里,209 条藏在同一作者的回复里,这些你自己刷 X 很容易漏掉。223 位创作者的署名和原帖链接都保留着,想看作者的其他作品,顺着链接点过去就行。
省掉翻找:按你要做的东西找。 17 个分类,电影叙事 131 条、音乐 MV 115 条、产品广告 101 条,还有时尚、动作、动漫、Vlog 等。要做什么,直接进对应分类挑。

省掉准备素材:大部分复制文字就能跑。 550 条里有 445 条是纯文生视频,不需要准备任何图片或视频,复制一段文字就能出片。
一条好提示词的结构,拆给你看
先回到开头那条锦鲤。完整提示词如下(原文照录,作者 @underwoodxie96):
15-second, 16:9 vertical, continuous single-take video that looks like authentic smartphone footage accidentally captured by a passerby in a city park. Overcast natural daylight, subtle handheld shake, limited phone stabilization, occasional autofocus adjustment, and realistic smartphone compression. The absurd event is filmed with a completely serious, unscripted documentary feeling. 0–3s: [Handheld medium shot] A middle-aged man wearing a dark business suit and tie crouches beside the stone edge of a pond. With a completely serious expression, he slowly scatters pieces of bread from a paper bag to several ordinary koi. Small ripples spread across the water as the fish gather in front of him. 3–7s: [Camera instinctively moves closer] An abnormally huge orange-and-white koi suddenly surges out of the murky water, briefly lifting its upper body above the surface and biting down on the entire bread bag in the man’s left hand. He freezes for half a second, then grips the bag with both hands and leans backward. Startled, the person filming steps back. The image briefly loses focus before locking onto the man and the giant fish again. 7–11s: [Close handheld action shot] The giant koi pulls violently toward the deeper part of the pond. The wet paper bag stretches, the man’s arms tense, and his leather shoes slide repeatedly across the wet stone. His knee strikes the edge of the pond. He tries to brace himself with his right foot, but the sole loses traction and his center of gravity moves past the edge. Water, pieces of bread, and fallen leaves scatter from the force as the camera operator hurriedly moves sideways. 11–15s: [Impact and final hold] The paper bag suddenly tears. The man loses all support and pitches forward into the pond, creating one heavy, realistic splash. The camera quickly tilts downward while keeping the center of the pond visible. The man resurfaces with duckweed covering his head and his wet tie stuck across his face. The giant koi calmly swims past him with the remains of the bread bag still in its mouth. The camera holds on the man’s stunned expression while the koi casually swims away beside him. Keep the man’s face, dark suit, tie, and paper bag visually consistent throughout. The koi must retain the same orange-and-white markings and enormous size. The pulling, sliding, loss of balance, and fall must show believable weight, inertia, traction, and water displacement. Natural park ambience and a realistic splash only. No dialogue, no subtitles, no music. Avoid cuts, character teleportation, changes in the fish’s size, extra limbs, and cartoonish acting.
很长,但拆开看,就是六件事:
- 先定谁在拍、用什么拍:路人用手机偶然拍到。手抖、对焦犹豫、压缩画质,全都写明。“像真的”就是从这里来的。
- 按秒写节拍:0–3 秒喂鱼,3–7 秒锦鲤出水,7–11 秒拉扯,11–15 秒落水。每段只发生一件事。
- 写镜头怎么动:方括号里的“镜头本能地凑近”“镜头快速下摇”,等于给摄影师的指令。
- 锁住不能变的东西:大叔的脸、西装、领带、纸袋,锦鲤的花纹和体型,从头到尾保持一致。
- 把声音写进去:只要公园环境声和落水声,不要台词、不要字幕、不要音乐。
- 点名要避免的翻车:不要剪切、不要角色瞬移、不要鱼忽大忽小、不要多出肢体。

第 6 条最容易被忽略,也最值钱。 AI 视频最常见的毛病,比如人物突然换脸、东西凭空消失,提前写出来,模型就有了“不许做什么”的边界。
如果你用了参考图或参考视频,仓库的说明里还建议加一条:每份素材负责什么。 比如图 1 管人物长相,视频 1 管动作和运镜。
再看两条:广告片和 Vlog
一条手机广告(作者 @UrMeer289)。 结构最简单的一类:三个时间段,每段写清楚画面和镜头,最后给出品牌名和 slogan。
Create a 15-second premium cinematic branding TVC for a futuristic smartphone brand called NOVA X.
0–5 seconds: Extreme close-up of a sleek futuristic smartphone resting on a glossy black surface. Dramatic studio lighting sweeps across the device, revealing its ultra-thin design and premium metallic edges. Cinematic camera push-in, macro product details, realistic reflections.
5–10 seconds: The phone activates with a stunning futuristic interface. A young professional picks it up, and the environment transforms into a vibrant digital world with floating holographic visuals, smooth UI animations, and seamless interaction. Dynamic camera movement, premium commercial aesthetic.
10–15 seconds: Hero product shot of the smartphone standing vertically against a minimalist futuristic background. The NOVA X logo appears elegantly with the tagline: “NOVA X — The Future in Your Hands.” Cinematic lighting, high-end TV commercial quality, photorealistic, ultra-detailed, smooth motion, polished advertising style, 4K.
NOVA X 手机广告(原作者成片),片尾品牌字和 slogan 都按提示词出现了
想做自己产品的广告,最省事的做法就是:保留三段式结构,把手机换成你的产品,把 NOVA X 换成你的品牌名。
一条健身房 Vlog(作者 @doctorwasif)。 这条的写法很像剧本:先用大写的 CAMERA、LOOK、STYLE、CHARACTER、SETTING 把设定交代完,再列出 6 个镜头,每个镜头一句台词。
CAMERA: DV 16mm tape camcorder POV. CHASE films herself, occasionally propping the camera on a wall or gym bag. Natural handheld shake, imperfect framing, delayed focus, clumsy zooms, face occasionally cut off. Camcorder never visible.
LOOK: Soft blurry tape aesthetic, faint tape noise, bloomed gym lights, flickering auto-exposure, muted contrast, realistic skin.
STYLE: Calm post-workout gym vlog. Relaxed, reflective, slightly tired but happy. Slow pacing, natural laughs.
CHARACTER: CHASE, Korean idol (20s). Long black hair loosely tied, light post-workout glow, slim athletic build. Modest long-sleeve athletic top, loose joggers/leggings, socks or shoes off, towel around neck, no jewelry.
SETTING: Quiet evening gym. Stretching mat, mirror wall, gym bag, protein shake, soft lighting, mostly empty.
15s | 6 Cuts
Propped medium: sits on mat, exhales. "Okay, cool-down time."
Handheld stretch: hamstring stretch, laughs. "Yeah, that was a lot today."
Macro: picks up protein shake, condensation visible. Ambient sound only.
Medium: drinks shake, smiles. "Okay, that's actually really good."
Close: overhead stretch, rates workout. "I'd give today like an eight out of ten."
Selfie: grabs bag and shake, waves. "Alright, I'm heading out—see you guys."
健身房 Vlog(原作者成片)
这种“设定 + 镜头清单”的写法,特别适合做 UGC 和口播类内容。 换个人设、换个场景、改几句台词,就是一条新视频。
怎么用:三种方式
方式一:直接抄。 打开仓库 README,每条都有视频预览和折叠的完整提示词。想按类型找,就进 prompts 目录,按分类、用途或分页翻;也可以看网页版 Gallery。看中哪条,复制、改主体,扔进 H3。
方式二:通过 API 批量跑。 如果你要一次出很多条,比如同一个产品出十个版本的广告,用 API 更省事。
MiniMax 官方有自己的 API,BeatAPI 上也能直接跑 H3。
用法很简单:一个 key,提交任务、轮询状态、拿视频链接,三步。不想写代码,也可以直接在 Playground 里把仓库的提示词粘进去跑。

大家最关心的两件事:多快、多少钱。
速度:平均约 45 秒,最快约 30 秒出片。
价格:按秒计费,最低每秒约 1 毛 2。 拿库里的提示词先用便宜的档位低成本试几版,挑出满意的,再用高清档出正片,是比较省钱的用法。
方式三:交给 Agent。 库里所有提示词都在一个 catalog.json 文件里,分类、时长、画幅、视频链接都是结构化的。你可以让 Agent 先读这个文件,按你的需求挑一条最接近的,改写后再通过 API 提交任务。
写在最后
这个库会持续更新,X 上新出的好提示词会陆续收进来。想第一时间看到,去 GitHub 点个星:
github.com/BeatAPI/awesome-minimax-h3-prompts
如果你也用 H3 跑出过好作品,欢迎投稿。仓库有现成的投稿表单,要求不多:完整的提示词、原帖链接和作者、参考素材分别负责什么,以及你有权公开的成片。只写“电影感、高清、爆款”的提示词不会被收录。我们更想要能教会别人一种拍法的提示词。
模型会越来越强,但“会写分镜”的人,永远先出片。
参考来源
- awesome-minimax-h3-prompts 仓库:https://github.com/BeatAPI/awesome-minimax-h3-prompts
- 仓库中文说明:https://github.com/BeatAPI/awesome-minimax-h3-prompts/blob/main/README.zh-CN.md
- 仓库投稿规则:https://github.com/BeatAPI/awesome-minimax-h3-prompts/blob/main/CONTRIBUTING.md
- 在线 Prompt Gallery:https://beatapi.io/zh/minimax-h3-prompts
- MiniMax 官方视频生成文档:https://platform.minimax.io/docs/guides/video-generation
- MiniMax H3 发布推文:https://x.com/MiniMax_AI/status/2083006198828417501
- 锦鲤案例原帖(@underwoodxie96):https://x.com/underwoodxie96/status/2082747838782386563
- NOVA X 案例原帖(@UrMeer289):https://x.com/UrMeer289/status/2083048872566575568
- 健身房 Vlog 案例原帖(@doctorwasif):https://x.com/doctorwasif/status/2083048782581858681
- BeatAPI MiniMax H3 API:https://beatapi.io/zh/minimax-h3-api