大家好,我是冷逸。
前几天,我们新造了一个IP,叫「沃」。今天拉他出来拍片子了,大家觉得怎么样?
喜欢的话,多多三连,以后争取让他多拍点片子。
你可能想问,这支短片怎么做的?还记得我们的lengyi-shotlist skill吗?这个片子就是用它配合MVLAND完成的。
lengyi-shotlist skill负责出Prompt,MVLAND负责设计资产图、生成视频以及最后的剪辑。
我还用它做了一支创意短片《门》。
刚开始,是妈妈抱着刚出生的女儿,第一次推门回家;多年后,变成女儿搀扶年迈的妈妈跨过门槛。爱在同一道门框里,完成了方向的对调。
保姆级教程
以上两支短片,都是用MVLAND做的。
他有3个地方,我觉得很有意思。
1)带剪辑台的画布
画布类视频工具,我们见过很多了,但MVLAND有自己不一样的地方。
我们以《门》这支短片为例,给大家详细看一下制作过程。
首先,用lengyi-shotlist skill设计一个完整故事线的分镜脚本。这个skill建议搭配DeepSeek v4 pro、GPT-5.5、Qwen3.7-Max、Opus 4.8这类推理+文学创作强的模型来用(我这次是用的Opus 4.8),在Claude Code、Codex、Workbuddy、TRAE Work、Qoder Work这类Agent里出提示词。
故事设定了十个开门瞬间。
skill会直出视频参考主体的I2V提示词和逐镜提示词。
然后,把提示词复制到MVLAND里,先生图,再生视频。
打开mvland.cn,创建一个画布。先出图,模型选"全能图片模型2.0"(其实就是Image2),调好清晰度和尺寸,点「生成」就行。
依样画葫芦,陆续得到了女儿和妈妈不同年龄段的图片。
接下来生成视频。和生图一样,选择参考主体,输入提示词,选择比例、时长和分辨率就可以了。
MVLAND的画布有一个比较有意思的设计:每个节点都有独立的历史管理能力。
比如我们生成了一段视频,不满意,需要重新生成。其他画布工具是把每一次的结果单独摆在画布上,MVLAND不是,它把所有抽卡结果都收在同一个节点里。
点左上角那个时间回滚的按钮,就能看到历次生成的视频,点击某个版本,就可以替换到主节点上。
所有视频都生成好后,换作其他画布产品,接下来你得挨个下载视频,再丢进剪映里剪辑。
MVLAND不用,它自带剪辑台,一键唤起。
剪辑台可以自由缩放,调整尺寸和显示比例,也可以全屏。
全屏界面,视频、字幕、音频轨道全都有,媒体池、时间线、素材编辑区等完整面板也都有,都能自由调整。
粗剪效果,不需要渲染就能预览。
剪好后,导出视频。
渲染速度还挺快的,我这个一分钟的短片,基本上十几秒就渲染好了。
另一支「沃」的短片,也是同样的制作路径。
不过它的逻辑更简单,直接4支15s视频拼接就行。
最终得到了这样一支短片。
整个体验下来,其他画布产品有的功能,MVLAND基本都有。但它还多了两个东西:多轨道剪辑和节点历史管理。
据我了解,MVLAND应该是目前市面上唯一同时支持这两个功能的产品。
2)一键制作MV
MVLAND还有两个有意思的功能:一键制作MV和一键做爆款。
这两个基本都是"懒人福音",就是一路点点点,就能出不错的作品。
比如MV制作,我随手上传一段之前做好的歌曲,它会先让我描述脑海里第一时间想到的画面。只需要填一些关键词就行,比如失恋、心碎。
不知道填什么?这里有个小建议:去qwen.ai问大模型。Qwen3.7-Plus原生支持多模态,可以帮你分析歌曲情绪。
填好关键词后,MVLAND开始分析。
它会给出一个详细的分析结果,包括曲风、节奏、艺术风格和MV角色,都可以自由选择。
我选了「魔幻史诗」风格,角色就是「沃」。
它会先生成场景图,再生成分镜图。我数了下,一共有47个分镜图。
如果对分镜图不满意,可以继续修改,重新生成。比如我分镜32这里有点小问题,我就让它重新生成了分镜图。
确认分镜图后,接下来就是调用Seedance 2.0、可灵数字人2.0等模型来生成视频。
所有生成好的单镜头、音轨和字幕,都会放进一个剪辑台,可以在里面直接剪辑、调整。
一起来欣赏下最终的MV吧。
这MV质量,确实很顶。
3)一键仿爆款
爆款玩法就更简单了。
MVLAND内置了很多爆款模板,有世界杯的、MV的、Vlog的,也有TVC的。
这回主角换成了庄颜同学,她最近拍戏累了,让她去罗马玩一下吧。
看到成片那一刻,这Vlog也太真了吧!
来源:沃垠AI
Hey everyone, I'm Cold.
A few days ago, we created a new IP called "Wo". Today, we pulled him out to shoot a video. What do you guys think?
If you like it, give me a lot of likes and shares so that I can try to get him to shoot more videos in the future.
You might be wondering how this short film was made? Do you remember our "lengyi-shotlist" skill? This film was completed using it in conjunction with MVLAND.
The "lengyi-shotlist" skill was responsible for generating the prompt, and MVLAND was responsible for designing the asset image, generating the video, and the final editing.
I also used it to create a creative short film called "Door".
At the start, it was a mom holding her newborn daughter, opening the door to go home for the first time; many years later, it turned into the daughter helping her elderly mom over the threshold. Love completed a direction reversal within the same door frame.
Step-by-step tutorial
The two short films above were both made using MVLAND.
There are 3 things about it that I find very interesting.
1) Canvas with an editing stage
We've seen a lot of canvas-based video tools, but MVLAND has its own unique features.
We'll use the "Door" short film as an example to show you the production process in detail.
First, use the "lengyi-shotlist" skill to design a complete storyboard for the story. This skill is best used with models like DeepSeek v4 pro, GPT-5.5, Qwen3.7-Max, or Opus 4.8, which are strong in reasoning and literary creation. I used Opus 4.8 this time.
The story sets up 10 moments of opening the door.
The skill directly outputs the I2V prompt for the video reference subject and the per-shot prompt.
Then, copy the prompt into MVLAND, generate the image first, and then the video.
Open mvland.cn, create a canvas. Choose the "Versatile Image Model 2.0" (which is actually Image2), adjust the clarity and size, and click "Generate".
Repeat the process to get images of the daughter and mom at different ages.
Next, generate the video. Just like generating images, select the reference subject, input the prompt, choose the aspect ratio, duration, and resolution.
The MVLAND canvas has a very interesting design: each node has independent history management capabilities.
For example, if you generate a video and aren't satisfied, you need to regenerate it. Other canvas tools would place each result separately on the canvas, but MVLAND collects all the attempts under the same node.
Click the time rollback button in the top-left corner, and you can see all the previously generated videos. Clicking on a specific version allows you to replace it with the main node.
Once all the videos are generated, in other products, you'd have to download them one by one and then import them into CapCut for editing.
With MVLAND, you don't need to, because it comes with its own editing stage, which you can activate with one click.
The editing stage is fully resizable, and you can adjust the display scale or even go full-screen.
In the full-screen interface, there are video, subtitle, and audio tracks, as well as a complete media pool, timeline, and material editing area.
You can preview the rough edit without needing to render.
After editing, export the video.
The render speed is surprisingly fast; my one-minute short film was rendered in about 10-20 seconds.
The other "Wo" short film was made using the same process, but the logic is simpler as it involves splicing together 4 15-second videos.
Overall, MVLAND has all the features of other canvas products, but it also adds two important features: multi-track editing and node history management.
To my knowledge, MVLAND is currently the only product that supports both of these features simultaneously.
2) One-click MV creation
MVLAND also has two interesting features: one-click MV creation and one-click trend following.
For example, for the MV, I uploaded a song, and it asked me to describe the images I imagined hearing it. I just wrote some keywords like "heartbreak" and "shattered heart".
Check out the final MV.
The quality is top-notch.
3) One-click trend following
The trend function is even simpler.
MVLAND has a lot of built-in popular templates for the World Cup, MVs, Vlogs, and TVCs.
Source: Wo Yin AI
来源:沃垠AI

























