频道:AI Samson · 时长:22:20
| 项目 | 内容 |
|---|---|
| 标题 | Seedance 2 Just Broke the Internet (50 Prompts You Have To See) |
| 教程编号 | 04 |
| 视频 ID | rXtuzDMLa_Y |
| 时长 | 22:20 |
| 频道 | AI Samson |
| 上传日期 | 2026-04-15 |
| URL | https://www.youtube.com/watch?v=rXtuzDMLa_Y |
| 主题 | Seedance 2 AI 视频生成模型:50 个创意提示词展示与实操教学 |
“Seedance 2 can create videos like this. Cool, isn’t it? And here are 50 incredible prompts that you can try out yourself. And you don’t need to be an expert. I’ll show you exactly how to create videos like these and give you all the prompts for free.”
Seedance 2 是最新发布的 AI 视频生成模型,它在互联网上引发了巨大轰动。这个模型能够生成长达 15 秒的高质量视频,并且拥有多模态输入能力——不仅可以用图片作为参考,还可以直接上传视频素材进行二次创作。更令人兴奋的是,它能原生生成与画面匹配的音频和对话。
AI Samson 在这期视频中精心收集了来自全球各地创作者的 50 个精彩提示词案例,涵盖了子弹时间特效、多镜头叙事、动画制作、3D 动画、广告创意、历史重现等多个领域。他不仅展示了这些令人惊叹的作品,还手把手教你如何使用 Higgsfield 平台来复现这些效果,并分享了自己的失败经验和调试技巧。
无论你是 AI 视频创作的新手还是老手,这 50 个提示词都为你打开了一扇通往全新创意世界的大门。接下来,让我们按照视频的节奏,逐阶段深入学习。
子弹时间(Bullet Time)是一种经典的电影特效——画面中的动作在某一瞬间被冻结,而摄像机围绕着冻结的场景进行环绕拍摄。这种效果最早在《黑客帝国》中大放异彩,而现在 Seedance 2 让普通人也能轻松实现。
“First up, we have this bullet time freeze in a beach volleyball match from Josh at altphotos_pl. We have this freezing moment in time where the camera orbits around the player before spiking the ball across the net.”
视频中展示了多位创作者的子弹时间作品:
“Jared Lu created his own version of this effect for a SWAT team entry with an actual bullet. So, here you can see someone burst in the door. And we get this wonderful orbiting of a frozen moment of time.”
AI Samson 还展示了自己创作的几个子弹时间变体:
“Another variation is slightly comical banana slip which shows a man, myself, mid-air falling with all of my documents on the floor before collapsing on the floor with a coffee on my face.”
Higgsfield 是本期视频的赞助商,也是运行 Seedance 2 模型的主要平台。以下是完整的操作流程:
“Well, how do you actually make this, you might be asking. Well, let me show you how. And for this we’ll be using Higgsfield who are also the sponsor of today’s video.”
“So we can come to the image tab inside of Higgsfield and here we can access a whole host of different image models. I’m going to be using the C Dream 5.0 light model.”
“Currently Higgsfield has to make sure that these images are eligible to be run inside of Seedance 2. So once you’ve uploaded them, they will be scanned and you’ll see they’ll come up as eligible or ineligible.”
这是 AI Samson 反复强调的核心技巧——不要从零开始写提示词,而是用现有的提示词模板,通过 LLM(如 Claude)来改写:
“So I suggest that you take the initial prompt if you’re changing it and come into an LLM. You can use any LLM. I like to use Claude at the moment… I’ve asked for the changes that I want as well as defining that we want to maintain the prompt structure as accurately as possible.”
AI Samson 诚实地展示了自己的失败案例,并分享了应对策略:
常见问题: - 视频前半段效果不错,但结尾动作完全混乱 - 人物的物理位置不合理(比如突然倒置) - 动作从一个位置突然跳到完全不同的位置
“Now, I’ll also show you a couple of failed attempts that I had myself… You’ll see that we have some odd physics here of the way that the man is located. You can see him here upside down, and then as he strikes the ball, suddenly he is in a completely different position.”
解决方案: 1. 简单重新生成:相同的提示词再跑一次,会得到不同的结果 2. 在 LLM 中重新编辑提示词:向 LLM 描述具体问题,让它在提示词中明确规避这些问题
“One is simply rerun the prompt, and you will get out a different result. The second is to re-edit the prompt inside of an LLM. So, what you can do is describe the issue to the LLM, and make it explicit that this is something we want to define in the prompt.”
除了子弹时间,Seedance 2 还能实现”暂停现实并探索它”的效果。
“There’s also this effect of pausing reality and exploring it. And here is an example from Okan Can. You can see him clicking his fingers while walking through a epic Roman battle, lighting a cigarette, and as he does so, allowing time to restart.”
Okan Can 的案例:一个人在史诗级罗马战场中打响指暂停时间,悠闲地点着一根烟,然后让时间重新流动。这个效果的提示词已经在视频描述中链接。
AI Samson 的变体:将场景改为第一次世界大战的战场。操作流程: 1. 在图片标签页生成自己在两军对峙中间的起始图 2. 用 Claude 修改提示词以适配新场景 3. 上传起始帧和更新后的提示词 4. 生成视频
“I’m going to change it to a World War I battle scene. So, I’m going to ask in the image tab to put myself in the middle of two armies from World War I charging towards each other.”
Seedance 2 的一个独特优势是其多模态能力——不仅支持图片输入,还支持视频输入。
“Seedance 2 gives us up to 15-second long videos, and one aspect that makes it entirely unique is its multimodal capacity. And that means that we can upload not just images, but also video.”
Elsina(X 平台) 展示了如何上传真实拍摄的视频素材,然后用 AI 生成不同的变体,包括改变场景和添加特效。
“I think this is one of the most exciting emerging techniques for using AI video. And that’s not replacing real-life footage, but it’s actually merging the two together.”
这是一个非常重要的观点:AI 视频不是要取代真实拍摄,而是将两者融合在一起。
多镜头提示词(Multi-Shot Prompting)是 Seedance 2(通过 SeaArt Suite 平台)的另一个强大功能。它允许你用一个提示词创建多个不同的连续镜头,同时保持角色外貌、场景位置和光照的一致性。
“Another incredibly powerful feature of SeaArt Suite is its multi-shot prompting capacity. And this means that with just one prompt we can create many different consistent shots of a situation whilst maintaining perfect character, location, and lighting consistency.”
1. AI with Sarah —— 快节奏动作序列 - 快速切换的多镜头序列,角色外貌始终保持一致 - 产生了真正沉浸式叙事的感觉
“We’ve got this fast-hitting sequence with many different shots whilst maintaining this character’s likeness… It creates a real sense of immersive storytelling.”
2. Tech Hala —— 变异人特效 - 一朵兰花从人脸上绽放的超现实场景 - 传达出强烈的紧张感和戏剧张力
3. Mr. Das(X 平台) —— 市场追逐戏 - 一个女人在市场中快速奔跑 - 所有镜头保持一致性,营造出速度感和沉浸感 - 关键技巧:他逐个描述每个镜头,精确命名每个镜头中的动作,完整的 15 镜头设置
“For the shot descriptions, he breaks down each shot individually and names exactly the action that’s going on. And we have this full 15-shot setup.”
4. Ivana —— 时装工作室 - 时装工作室更衣室场景,不同模特准备走秀 - 镜头从每个角色依次扫过,最后所有人一起走向舞台 - 以黑屏结尾的剪辑效果令人赞赏
但 AI Samson 也诚实指出了瑕疵:
“You will notice here this woman pushing away the clothing rail and it somewhat morphing into the background. These are the type of things that are still going to apply reality and incredibly difficult to get rid of.”
5. umesh_ai —— 无人机跟踪镜头 - 快速无人机镜头,从远处俯冲到山顶上站立的人物 - 镜头的流畅度令人惊叹 - 这种动态镜头在现实中需要极其昂贵的设备才能实现
“What’s wonderful about this is the smoothness of the camera and how we’re getting this dynamic challenging shot that would require extraordinary equipment and capacity to film using a drone.”
6. sanyaspeaks —— 广告级早餐流程 - 展示了如何用多镜头提示词开发广告级内容 - 一个人的早餐日常,通过精美的剪辑展示最有趣的部分
7. VR 眼镜场景 —— 沉浸式叙事 - 一个人在北极荒野中与北极熊搏斗,然后突然被拉回现实,摘下 VR 眼镜 - 特别精彩的细节:她在 VR 中双手捂脸的动作被完美映射到回到现实的镜头中
“One is how she puts her hands to her face and that being mimicked into the shot as we return into reality.”
AI Samson 自己也尝试了这个效果:
“So, here’s me dueling a polar bear. Samson! What is it? Pretty impressive.”
8. Iqra Saifi —— 鲸鱼上的芭蕾舞者 - 一位芭蕾舞者在巨鲸背上跳舞的超现实镜头 - AI 已经能创造出足够遵循物理规律、让人沉浸其中的超现实特效
“We’re getting to this point where we can create complex surreal special effects that still maintain the laws of physics enough that it become interesting and immersive for us to watch.”
AI Samson 也忍不住亲自上阵体验了在鲸鱼上跳芭蕾的感觉:
“There I am doing a wonderful leap, a fine pose, and a little pirouette with all of these fish flying around me. Very elegant and graceful.”
Tech Hala 展示了一个塔伦蒂诺(Tarantino)风格的多镜头电影级序列,角色演技出色,光照和场景之间保持了出色的一致性。
Seedance 2 的一个极其重要的特性是原生生成音频和对话。
“Another important factor about Seedance 2 is that it’s generating audio and dialogue natively.”
Alate(X 平台) 的舞蹈案例完美展示了这一点:
“Just notice how well the movements of the dancer match up to the audio. Particularly look at how well it matches this accent in the music to the foot clap here.”
舞者的每一个动作都精准地与生成的音乐节拍同步——甚至脚步的拍击声都恰好对准了音乐的重音。
Seedance 2 能创造出在现实中需要花费数万美元才能实现的特效:
“Seedance 2 gives us the possibility to create things that would cost tens of thousands of dollars to create in real life.”
案例展示:
“I adore these tracking shots that we’re getting out of Seedance 2. There is a lovely use of accelerating decelerating time to accentuate the movement of different objects.”
加速和减速时间的运用是 Seedance 2 的标志性美学之一。
Elsini 展示了如何用 Seedance 2 制作高质量的鞋类广告,但 AI Samson 也指出了不足:
“There was quite a lot of morphing going on on the face here. She looks a little bit skeletal. And you can also see there being a lack of detail in many of the finer elements. For example, there is a lot of disintegration of the quality on the text here.”
统一标签系统(Unified Tagging System) 是使这些复杂场景成为可能的关键技术:
“What makes this possible is the unified tagging system, and that means that you can reference images you upload as variables when we’re building out our prompts.”
在 Higgsfield 中使用的方法: 1. 在提示词中输入 @** 符号 2. 选择你之前上传的特定素材 3. 系统会自动理解你引用的是什么 4. 你还可以保存元素**,以便在不同作品中重复使用
“All you have to do is add in @, and then you’ll see you’ll be able to select the specific asset that you want to mention.”
Umesh(X 平台) —— 战场上飞行的箭矢: - 从宏观战场镜头推进到箭头特写,展示细腻的雕刻纹理 - 从宏观到微距的镜头过渡能力令人惊叹
第一人称冻结时间: - 第一人称视角在冻结的时间中穿行 - 这种复杂的电影摄影技巧在没有 AI 的情况下很难实现 - 为游戏创意和沉浸式内容开辟了新的可能
“Getting this effect gives us the potential to be generating complex game ideas as well as extremely immersive content.”
Keda Ala —— 超级英雄变身:角色快速变身为超级英雄版本
I am Emily 2050 —— 猫咪救援场景: - 一个女人在雨中跳向一只猫,阻止标志砸到它头上 - 亮点:湿漉漉地面上溅起的水花,以及女主角优雅的扑救动作
她使用了一种特别的提示词技术——关键物体规则(Critical Object Rules):
“You’ll see the prompting technique that she is using here to define each of the shots and how long they will last as well as critical object rules. So, these are parts that help to get very specific moments inside of the sequence.”
AI Samson 建议:如果你在创作中遇到间歇性问题,可以添加这类关键物体规则来约束特定元素。
Seedance 2 不仅擅长写实视频,在动画内容方面同样出色。
“It’s not just real-life footage that Seedance 2 is handling with ease. It’s also doing particularly well with animated content.”
AI Gezgini —— 动画风格的双人战斗序列
El Cine —— 草地环境中的飞行穿越: - 主角在草地般的环境中穿梭,加速减速的节奏感极佳 - 最终降落在一只狗的鼻子上 - 使用了 JSON 格式的提示词结构
“He is using a JSON prompt formatting structure, which can allow you to get much more specific elements. Now, if you want to use a JSON prompting structure, you can simply take your prompt and ask an LLM to structure it in a JSON format.”
Tom likes robots —— 动漫风格: - 画面非常精致细腻,有着吉卜力工作室(Studio Ghibli)的感觉 - 已经达到了可以专业使用的水平
“The artwork here is very delicate and precise. And it certainly has the feel of a Studio Ghibli film. And you would have to say that this is at the level where it could be used professionally.”
AI Samson 提出了一个重要观点:真实感视频虽然令人印象深刻,但往往能看出是 AI 生成的;而艺术风格的动画内容则更难被识别为 AI 作品。
“For some of the real-life videos that are coming out, there is a huge degree of impressiveness, but you can often notice evidently that it’s AI-generated. Whereas with this type of artistic content, it is much harder to notice anything that would make you think it was generated by AI.”
Amira Zaeiri —— 小巷中两个角色对话的动画场景,展示了创作长篇动画内容的可能性。
AI Samson 提到了一个值得关注的现象:AI 生成的病毒式 LEGO 视频正在被用作宣传工具,获得了数百万的观看量。
“There are viral LEGO videos created with AI that are being used as propaganda, and these are gaining millions of views. And it demonstrates the capacity for AI-generated media to have real impact on culture and politics.”
Latte(OXBisk,X 平台) —— 被蛇追逐的小男孩: - 在亚洲城市街道中奔跑的小男孩被蛇追逐 - 有着皮克斯(Pixar)电影般活泼的动画感觉
“It suddenly has the playful animated feel of a Pixar film.”
“Some of the most interesting prompts that I found revolve around creating 3D animations.”
Simply Anissa —— 伊斯坦布尔纸面地图 3D 化: - 伊斯坦布尔的地图从纸面上立体浮现 - 美丽的建筑群从平面中生长出来 - 非常适合作为片头序列使用,甚至让人联想到《权力的游戏》的片头
“You can see just how well this could work for a title sequence. It even reminds me a little bit of the title sequence of Game of Thrones.”
视频中展示了伊斯坦布尔大围城的历史飞越镜头,用沉浸式的方式重现历史。
“The possibility of recreating history is an extremely immersive way to learn about our past.”
AI Samson 特别提到了一个成功案例 —— Majestic Studios 频道: - 已拥有 189,000 订阅者 - 单个视频获得 40 万以上观看量 - 完全使用 AI 生成的长篇历史内容 - 展示了曼彻斯特的历史变迁
“This channel Majestic Studios already has 189,000 subscribers and 400k views on this video alone on a long form generated piece of content from AI.”
即便内容并不完美(比如马匹在田野上行走的方式看起来很奇怪),但依然能够吸引大量观众。
“Even this content is not perfect at all. You can see how weirdly these horses are walking across the fields. It’s still interesting enough to gather a huge audience.”
特洛伊木马历史飞越:古希腊特洛伊木马的电影级展示,将历史或神话的真实时刻以沉浸式的方式呈现。
“I can certainly see there being a huge market for people recreating specific eras and stories from our past with AI.”
| 概念 | 说明 |
|---|---|
| Seedance 2 | 最新 AI 视频生成模型,支持多模态输入,原生音频生成,最长 15 秒 |
| Higgsfield | 运行 Seedance 2 的主要平台,提供图片生成和视频生成工具 |
| C Dream 5.0 Light | Higgsfield 平台上的图片生成模型,用于创建视频起始帧 |
| Bullet Time Freeze | 子弹时间冻结效果,画面冻结后摄像机环绕拍摄 |
| Multi-Shot Prompting | 多镜头提示词,用一个提示词生成多个一致性镜头 |
| Unified Tagging System | 统一标签系统,用 @ 引用上传的素材作为提示词中的变量 |
| JSON Prompt Structure | JSON 格式的提示词结构,允许更精确地控制各个元素 |
| Critical Object Rules | 关键物体规则,在提示词中约束特定物体的行为 |
| Multimodal Input | 多模态输入,支持图片和视频作为参考素材 |
| Native Audio Generation | 原生音频生成,AI 自动创建与视频匹配的音效和对话 |
不要从零写提示词 —— 始终以已验证的提示词模板为基础,通过 LLM(如 Claude)进行改写。这样可以保持有效的提示词结构,同时适配你的创意。
用起始图片定义场景 —— 在生成视频前,先用 Higgsfield 的图片功能创建一张高质量的起始帧。上传参考图(灵感来源)和你自己的照片进行合成。
失败是正常的 —— 即使是经验丰富的创作者也需要多次尝试。遇到失败时,要么简单重新生成,要么在 LLM 中精细化提示词来描述并解决问题。
逐镜头描述多镜头序列 —— 在使用 Multi-Shot Prompting 时,像 Mr. Das 那样逐个定义每个镜头的具体动作,这样能获得更好的叙事连贯性。
善用 @ 标签系统 —— 在 Higgsfield 中通过 @ 符号引用你上传的素材,让 AI 精确理解你指的是哪个角色或物体。保存常用元素以便复用。
动画风格更容易出彩 —— 写实视频容易暴露 AI 的痕迹,而动画和艺术风格的内容更难被识别为 AI 生成,也更容易达到专业水准。
尝试 JSON 格式提示词 —— 如果需要更精确的控制,可以让 LLM 将你的提示词转换为 JSON 结构格式。
添加关键物体规则 —— 当你遇到间歇性的物理或逻辑问题时,在提示词中添加 Critical Object Rules 来约束特定元素的行为和位置。
误以为必须是专家才能使用 —— AI Samson 明确说”you don’t need to be an expert”,只需使用现成的提示词模板即可上手。
从零开始写提示词 —— 这是最常见的错误。应该以已验证的模板为基础,通过 LLM 进行修改,保持有效的提示词结构。
期望一次就能得到完美结果 —— 即使经验丰富的创作者也有大量失败尝试。AI Samson 自己就展示了多个失败案例。正确心态是迭代优化。
认为 AI 视频是要取代真实拍摄 —— AI Samson 强调最激动人心的方向是将 AI 和真实素材融合使用,而不是替代。
忽视起始图片的重要性 —— 很多人直接用纯文本提示词生成视频,但使用高质量的起始帧图片能大幅提升视频效果和角色一致性。
认为 AI 动画不够专业 —— 动漫和艺术风格的 AI 视频已经达到了”could be used professionally”的水平,特别是吉卜力风格的作品。
忽略音频同步能力 —— Seedance 2 的原生音频生成不是简单的背景音乐叠加,而是与画面动作精确同步的智能音频。
低估 AI 历史内容的商业价值 —— Majestic Studios 的案例证明,即使内容不完美(马的动作不自然),AI 历史频道也能获得 18.9 万订阅和 40 万单视频播放量。
不检查素材兼容性 —— 在 Higgsfield 中上传图片后,必须等待系统扫描确认是否 eligible,否则无法使用 Seedance 2 生成。
忽视多镜头提示词中的详细描述 —— Multi-Shot Prompting 的效果很大程度上取决于你是否为每个镜头提供了详细的独立描述。
Seedance 2 支持最长 15 秒的视频生成,这在 AI 视频领域是一个显著的进步。
多模态输入是 Seedance 2 的核心竞争力 —— 同时支持图片和视频作为参考输入,这使得”真实素材 + AI 增强”的工作流成为可能。
多镜头提示词(Multi-Shot Prompting)实现了叙事级别的 AI 视频,一个提示词即可生成保持角色、场景和光照一致性的多镜头序列。
原生音频生成让 AI 视频更加完整 —— 不只是画面,音效和对话也与画面动作精确同步。
提示词模板 + LLM 改写是最高效的创作方法 —— 不要从零开始,站在前人的肩膀上进行创新。
统一标签系统(@)使复杂场景成为可能 —— 通过变量引用上传的素材,实现精确的元素控制。
动画和艺术风格的 AI 视频已达到专业水准 —— 特别是动漫风格,比写实视频更难被识别为 AI 生成。
AI 历史重现内容具有巨大的商业潜力 —— 已有频道仅靠 AI 生成的历史内容就获得了十几万订阅者。
失败和迭代是 AI 视频创作的常态 —— 物理错误、动作混乱等问题很常见,但可以通过重新生成或精细化提示词来解决。
AI 视频在广告领域有巨大应用空间 —— 可以创建成本仅为传统制作零头的高质量广告和产品展示视频。
JSON 格式提示词可以获得更精确的控制 —— 当标准提示词不够精确时,尝试用 JSON 结构来组织你的提示词。
关键物体规则是解决物理和逻辑问题的有效工具 —— 在提示词中明确约束特定物体的行为,可以减少不合理的生成结果。
Seedance 2 的发布标志着 AI 视频生成进入了一个新阶段。它不仅在画面质量上有了显著提升,更重要的是通过多模态输入、多镜头提示词和原生音频生成等功能,让创作者拥有了前所未有的控制力和创作自由度。
从子弹时间冻结到暂停现实,从动画短片到历史重现,从广告创意到游戏概念——这 50 个提示词案例展示了 AI 视频创作的广阔前景。无论是个人创作者还是专业团队,都能从这些技术中找到适合自己的应用场景。
正如 AI Samson 在视频中所展示的,即使是经验丰富的创作者也会遇到失败,关键在于掌握正确的方法论:以模板为基础、用 LLM 优化提示词、通过迭代不断改进。这个方法论比任何单一的提示词都更有价值。
“Seedance 2 is a huge step forward in the power for AI video, giving us greater control over characters, locations, and lighting consistency. Its multimodal capacity allows us to use not only images, but also video as references. Seedance 2 has the capacity to create authentic intelligent audio that realistically matches our video. And it’s creating new possibilities for creating meaningful media either for educational or entertainment purposes.”
AI 视频的未来已经到来,而你现在就可以开始尝试。