AIGC trend sweeps TikTok, AI short video creation guidelines are here

Source: Dewai No. 5

Author: Junyi Yu Haijiao

How to use generative AI to assist short video creation, here is a guide.

AIGC's creative wave is sweeping TikTok. After the launch of image-based AI generators such as Stable Diffusion and Midjourney, a new wave of creative creators have used such technologies to create a series of "rewriting history" content that combines historical figures and events with modern elements and popular language to show New ideas and perspectives. With such content, some accounts have hundreds of thousands of followers. How to use generative AI to create short video hits? Let's take a look at TikToker's actual combat experience.

Legend: A short video on the theme of "What if Somalia conquered Europe?" created using AI (Source: TikTok)

Generative AI Empowering Short Video: Filling the Blank of Imagination

TikTok account @what.if_ai released a short video "What if India rules Britain?", which has been viewed more than 1.7 million times. The video tells a fictional backstory: the industrial revolution leading to the full militarization of India, the Indian fleet taking control of Britain, and wealthy Indian families migrating and settling in Europe. Queen Elizabeth married into the Indian royal family in gorgeous traditional Indian attire.

The @what.if_ai account operators claim that **most of their ideas come from the comment section. After studying the historical background, they used ChatGPT to generate scripts, and used Midjourney to make corresponding visualization effects, and finally made a video by means of image carousel. **Before the emergence of generative AI, such a short video account may need to form an operation team, including historical researchers, screenwriters, illustrators and video editors. But, "Now like me, no tech noob, just imagination, can create short video content with millions of views...Anyone can do it, I'm just an early adopter Early adopters."

Nayana Prakash, a PhD at the Oxford Internet Institute, said videos like this one took off because they filled a void in people's imaginations. **

"If the Mayan Empire never perished," she discovered, was the title of a short video that went viral on TikTok, showing pictures of Mayan-influenced cityscapes and magnificent buildings, all with AI Generated. It is worth mentioning that with the help of technology, modern skyscrapers can travel through time and space and be reinterpreted with ancient aesthetics.

Legend: Screenshot of the short video on the theme of "If the Mayan Empire never perished" (Source: TikTok)

** This wave of "rewriting history" is being staged on Western social media. **On the YouTube platform, some 20-minute-long videos explore and imagine some social issues more deeply, such as what kind of chain reaction will be produced if the losers of the world war win. These types of videos often make use of extensive slideshows, archival photos, and annotated maps to appeal to users interested in historical topics. On TikTok, the audience is wider, and historical works of this type of AIGC bring more users into the virtual world.

Generative AI short video creation guide

So, how to use ChatGPT and Midjourney to create short video works? Here is a practical guide.

Step 1 Use ChatGPT to create Midjourney hints

To generate an image on Midjourney, you first need to tell it what image to generate. Prompt words are inseparable here. Prompts () are a language used to interact with AI models, the purpose of which is to tell the model what content or images to generate. In the field of natural language processing, a cue word is usually a sentence used to describe a problem, task or situation.

How to generate prompt words that Midjourney can understand and output corresponding images, ChatGPT is an efficient assistant, which not only simplifies the workflow, but also saves time and energy.

**But as an intelligent robot, ChatGPT first needs to understand the necessary information. **

In this example, I'll start by telling ChatGPT the purpose: "I'm going to create the AI graph, but I'll give you the information first." I then give ChatGPT two pieces of information: how Midjourney works and its parameters In summary. This process will help ChatGPT understand how to compose Midjourney prompts.

Legend: Feed the basic information of Midjourney to Chat GPT (source: metaroids.com)

In the process of "educating" ChatGPT, it may give you a direct response. It is recommended to add a simple note at the end of the Midjourney information, telling ChatGPT that you don't want to receive a lengthy reply, just reply OK to express your understanding of the information.

Midjourney's Guide to Cue Words:

Step 2 provides prompt word samples for the next creation of ChatGPT

At this point, **ChatGPT already has some idea of how Midjourney works, but there is still one thing missing - a sample. **We need to tell ChatGPT how other creators use prompt words to create amazing images. Here is a commonly used prompt word formula to help ChatGPT generate prompt word content faster.

Legend: Feed the sample to ChatGPT (source: metaroids.com)

Prompt words: detailed description of screen theme + camera settings + Midjourney parameters

In terms of image generation, detailed descriptions of pictorial subjects typically include painting objects, object descriptors, and stylistic modifiers. These hints can help the AI model understand the type of image and generate a work that meets expectations.

**Painting objects:**Subjects of the picture, such as girls, tigers, sky, etc. For example, if you want to create a landscape painting, then the painting objects may include natural elements such as mountains, water, trees, etc. These objects are at the heart of the generated image, as they determine the composition and theme of the entire picture.

Object descriptors: are used to describe the attributes and characteristics of the painting object more specifically. For example, when describing mountains, you can use words such as towering, steep, and desolate to describe their characteristics. These descriptors are critical to ensuring that the AI model understands the type of image you are expecting.

Style modifiers: Used to describe the image style. Different style modifiers allow the AI to create more unique artwork. For example, if you want to create an Impressionist painting, descriptors such as brightly colored and roughly brushed would be good choices.

The third step is to give ChatGPT a role to make it generate more appropriate prompt words

In the previous two steps, ChatGPT has demonstrated its extraordinary ability to understand massive amounts of information, and it is like a student eager to keep learning. But before getting the prompt words generated by ChatGPT, you need to tell ChatGPT what role it should assume, so as to provide a more appropriate recovery.

For example: "You are a professional photographer. I want you to provide me with descriptive cue words that I can use for midjourney, including parameters and camera settings. Other than that, no other details are required."

Based on this, ChatGPT can become more "professional", so as to gain an in-depth understanding of professional knowledge related to photography.

Once you have a character, you can give it portrait ideas, for example: "Give me a cue for a photo of a futuristic live band concert. Please create a cue based on the sample format above. Be careful not to copy the cue above , for format reference only."

You can directly copy the hint words provided by ChatGPT. But you can also modify the optimization prompt words yourself if you want to get more varied results.

Although the image effect generated by Midjourney is random each time, you can let AI create similar layouts, scenes, colors, etc. by adding reference images. Adjusting the description, scale parameters, or providing a reference image often results in a more accurate image.

Step 4 Put the prompt words created by ChatGPT in Midjourney

ChatGPT will generate prompts based on the concepts you provide. Next, you only need to visit the Midjourney channel on Discord, enter /imagine in the dialog box, paste the prompts generated by AI, and you will see the desired image appear in front of you.

In order to help you digest the knowledge points of the above guide, let's look at three more practical cases:

Case 1: Elon Musk as a futuristic hero

Prompt Words (In order to make it easier for everyone to understand, we have translated the English prompt words into Chinese. But at present, Midjourney is more accurate in recognizing English prompts): "Take Elon Musk as a futuristic superhero in his SpaceX headquarters. Reimagining portraits. Musk should be confident, posing strong, and wearing a custom space suit that emphasizes the image of a superhero. The background shows SpaceX's advanced technology and mechanical equipment, focusing on futuristic elements. Use continuous autofocus. Mirrorless camera captures intricate mechanical details in the space suit and background. Set ISO to 800 and shutter speed to 1/200 for a balanced exposure. Using a wide aperture of f/2.8 to create a shallow depth of field, the Musk isolates from the background. The resulting image should be high resolution and cinematic, with color grading that complements the futuristic theme. Version v5."

Legend: Midjourney generates pictures

Warm reminder, the prompt words generated by ChatGPT sometimes encounter situations that Midjourney cannot recognize. The solution is to delete or modify the unrecognized content.

Case 2: Realistic artwork

Cue Words: "A beautiful, realistic work of art featuring a smiling golden retriever standing in a field of green grass and wildflowers. The puppy has soft, fluffy fur with individual strands Clearly visible. The background should be blurred, focusing on the characteristics of the dog. Use the v5 version."

Legend: Midjourney generates pictures

Case Three: Integrating into the Ocean

Prompt words: "Produce a surreal image of a mysterious and unknown subsea ocean. Use a wide-angle lens to create a sense of depth and scale, and set the aperture to f/11 to ensure deep depth of field. Set the shutter speed to 1/ 200 seconds to capture sharp details, and set the ISO to 400 to reduce noise in low light conditions. Try using long exposures to create a dreamy effect on water and moving creatures. Adjust the colors to capture the essence of the scene Mystical and otherworldly, such as adding a cool blue tint, or adding a soft glow effect to enhance the surreal character of the image. Take v5."

Legend: Midjourney generates pictures

Conclusion

With the rapid development of science and technology, the way of information dissemination is changing with each passing day. Generative AI has shown amazing capabilities in empowering the imagination of media people, extending the scope of content expression, and innovating expression methods. Under the premise of legal compliance, media professionals need to actively embrace new technologies to promote the development of media integration.