We have seen a lot of AI pictures, so have you seen AI "filming"? With the free opening of Runway's Gen-2, the threshold for AI video generation has dropped a lot more than expected.

After reading this article, you can even recreate the dream you dreamed last night with AI video. There is no doubt that you can become "Zhang Yimou" by making good use of AI.

Vincent Video | Say it, it will act

"Text generation video" is the most basic function of Gen-2. It is similar to the familiar AI drawing such as Stable Diffusion and Haoshuo painting. Just input text and Gen-2 can generate a video that matches your description.

However, it should be noted that Gen-2, like SD, cannot recognize Chinese prompts very well, but through the shortcut command function of Haoshuo AI (click " Read the original text " to download), the AI ​​language model can be used to accurately and quickly The Chinese description is converted into English Prompts to achieve our expected effect:

Prompts optimized shortcuts:

https://link.haoshuo.com/s/rygx4l

Regarding the basic use of "Text Generation Video", we have also made a short tutorial video before. You can follow the short video and input the English Prompts converted by Haoshuo AI to get the desired video:

Tusheng video | Let the picture move

Of course, "text generation video" is only the most basic usage of Gen-2. It is full of instability just like SD's "Vincent diagram", and the chance of "drawing crookedly" is not small.

However, SD has a "picture-generated picture", and Gen-2 also has a reference mode that supports uploading pictures. If we use AI tools such as painting and drawing to generate suitable reference images in advance, and then feed them to Gen-2, then Gen-2 will have a benchmark in the process of generating videos, at least the screen will not deviate too far.

It is easy to say that painting is very simple to use, and it usually takes only a few steps to generate a suitable picture:

Then we save the generated picture locally, upload it to Gen-2, and still input the English prompt words converted by AI,

Click "Free Preview" to preview the effect for free, and click "Generate" to wait for the result:

Tusheng video | sketches can also become big

Sometimes, we want to make this "AI blockbuster" more personal. At this time, we can take pictures of our hand-painted drafts and upload them to the advanced panel of Haoshuo Painting → Tusheng In the picture, add the descriptor:

Suggestions for parameter setting of "Graph Generated Graph"

01 It is recommended to use the shortcut commands mentioned in the "Wen Sheng Video" section to optimize Prompts

02 For the "drawing picture" of hand-drawn sketches, please increase the redrawing range to above 0.76

03 Since the material will be used in Gen-2, please turn off the "Show original image" option

04 Model and clipping mode can be set according to your preferences

Let the AI ​​"polish" it, and then follow the above-mentioned "picture video" steps to operate, and a short film from your "pen" will be generated:

AI makes blockbuster movies | Even screenwriters are saved

After mastering the two basic applications of Gen-2, we will finally introduce an advanced application using Gen-2: using AI to fictionally "shoot" a short film.

First, let’s take a look at the "Barbieheimer" movie trailer produced by netizens using Gen-2 to make full use of it-a fan AI short film that combines the themes of "Barbie" and "Oppenheimer":

Doesn't it look like it? It even makes people want to go to Maoyan to search for this film... Of course, this is the result of hard training by the bosses, but after reading this article, we can still find some practical methods.

First of all, to make a movie requires us to have a "content" plot and storyboard. Fortunately, we have AI that can learn from others' strong points and make brilliant writing. We will tell Haoshuo AI the theme we want to do, and use the shortcut command below to quickly generate 4~5 English Prompts for the story:

example

In this article, we will take " a little girl was singing on stage, and a boy fell in love with her at first sight, and he confessed to her on stage " as an example to demonstrate.

Prompts split-shot generation shortcuts:

https://www.hayo.com/prompt/19871

Then, we will generate the corresponding reference images in Haoshuo Painting one by one according to these storyboard prompts:

Example: Generated 4 storyboard reference pictures

Then use these reference images and the corresponding English prompts to generate videos according to the steps of "Generating Videos from Images":

It sounds like a fool's operation, but comprehending the following knowledge points and advanced skills is the key to success.

knowledge points

free limit

Gen-2 provides free users with a credit of 125 points. It supports Gen-1/Gen-2 to generate up to 4 seconds of video content at a time. There are restrictions on video resolution and material capacity, and cannot be removed. watermark.

number of seeds

Like SD, Gen-2 also has the concept of "seed number". By fixing the number of seeds, a relatively consistent picture look and feel can be maintained.

insert frame

Similar to SD's "HD Repair" function, but what it repairs is not the definition, but the fluency. After turning it on, you can get a smoother video, but the required generation time will also increase.

advanced skills

Although we can only generate 4 seconds of video at a time, some big guys found that as long as we fix the number of seeds, use the last frame of the generated video as a new reference image, and match it with the original Prompts, we can continue to generate coherent images. sexual image.

Through this little trick, as long as you have enough credit and patience, you can generate a video of tens of seconds or even several minutes.

In the end, if we use software such as Shangyingying for a little post-production, a romantic short film of "love at first sight on stage" will be completed:

That's right! Use your imagination and make good use of AI tools, and you will be the next "famous director"!