GTA 6 finally has its first trailer! Rockstar Games released the long-awaited “Grand Theft Auto VI Trailer 1” via YouTube on December 5th, 2023. The official GTA VI trailer was released hours early following a confirmed leak. Thankfully, fans are firmly focused on the trailer, its quality, and the newly revealed details. With the GTA 6 trailer released amidst the global consumer AI boom, AI-generated art is swiftly emerging based on the latest installment to the gaming franchise. Here’s how you too can use generative AI technology to create your own GTA 6 artwork!
We trained AI on Grand Theft Auto VI Trailer 1
With the trailer released plenty of time before the game itself — more than a year in advance — AI-generated images are all we’ve got for now. We generated the following AI images based on frames of the new GTA 6 trailer.
Source: Grand Theft Auto VI Trailer 1
Source: Midjourney
Source: Grand Theft Auto VI Trailer 1
Source: Midjourney
How to create AI GTA 6 images
Here’s how to use Midjourney to ‘train an AI on your own images’. Technically, this isn’t the real definition of “training an AI”. Doing that is much more complicated — so here’s the accessible version you can try at home!
Step
Download Discord
Firstly, download Discord for desktop or mobile (iOS or Android).
It can be accessed via web browser, but we recommend the standalone software for this purpose.
You’ll need a Discord account (which can be free), and a paid subscription to Midjourney to proceed.
Step
Install Midjourney
Install the Midjourney bot within Discord. Midjourney offers a comprehensive tutorial on this fairly simply process.
After this, you’ll be able to use Midjourney to generate AI images, within the public Midjourney channel.
We advise you set up your own Discord server, which is free and easy to do. This will keep your work organised, and stop your generations from scrolling up the screen as thousands of other users use the public channels.
Step
Write your prompt
Write your prompt!
Your prompt may look like “GTA 6, videogame sunset, pink and orange, busy highway bridge over a river –v 5.2 –ar 16:9”
This recreates the opening cinematic shot of Grand Theft Auto VI Trailer 1.
You can tell Midjourney to base your generated image on a existing image by putting a link to that image preceding everything else in the prompt. If you’d like to try this out, proceed to the next step!
In this example, the link would come before “GTA 6” and a comma is not required after the link. Also in the prompt above, suffixes like v (which stands for version) and ar (which stands for aspect ratio) are preceded by two hyphens. WordPress may display two adjacent hyphens as a single line, so you may not see it as two lines above. However, you must use two hyphens with no space between them for suffixes in your prompt.
Step
Image-to-image with Midjourney
Provide an input image for Midjourney to base your generated image on. You can do this in two ways. Either:
- Drag and drop an image from your computer into the prompt window.
- Screenshot something and paste it with the prompt window selected.
Either way, submit it so that Discord has it saved. Then, left click to maximize the image, and right-click the maximized image. Now click “Copy link” and you’ll have the link you need to use at the start of your prompt!
Midjourney doesn’t strictly do image-to-image generation, but it does do “image and text”-to-image. Some amount of text prompt is required along with your input image, even if that text is a single character.
Essential AI Tools
Content Guardian – AI Content Checker – One-click, Eight Checks
Jasper AI
WordAI
Copy.ai
Writesonic
Is GTA 6 going to have AI?
It’s all but confirmed that GTA 6 will use AI — in a particular way, and the distinction is very important. In fact, it’s probable that the next installment of the Assassin’s Creed, Red Dead Redemption, and Elder Scrolls franchises will all follow suit. The kind of character-based AI we’re referring to applies to any open-world game. We’ve already had one major GTA 6 leak, causing at least $5 million in damages to Rockstar Games and landing the perpetrator in a hospital prison until they promise not to repeat the offense. The hacker was an 18-year-old member of the international hacking group Lapsus$. While there’s no new info to confirm until someone new dares to leak it, I can at least give my personal prediction; Watch this.
My personal prediction
All future games in the Grand Theft Auto, Assassin’s Creed, Red Dead Redemption, and The Elder Scrolls franchises will use a form of AI.
Here’s the thing — this isn’t a contrarian stance. I’m no Nostradamus. Video games have been using a form of AI for years, and that form is NPC pathfinding. It’s possible to code a virtual brain that allows in-game non-playable characters (NPCs) to find their way around a 3D environment, even if that environment is made after the code. When you think about it, it’s obvious that the developers working on these open-world games aren’t programming the exact path of each and every NPC. All they know is where the floor is, and that they can’t fall through it. Now define an area, and they’ll explore it!
Beyond that, the next most obvious form of AI for open-world games is NPC dialogue. This could mean text-based dialogue or fully voiced AI-generated audio dialogue — both of which have reached a level of maturity this year that makes their use in a game like Grand Theft Auto 6 feasible.
Not all AI results in a digital media file, such as an image. Some generative AI creates instructions for your computer, your software, your mouse cursor, or an in-game NPC to follow. These instructions are generated, but not saved anywhere as a file (except temporarily within RAM and your CPU cache, if we’re being picky), and instead are immediately executed. In fact, not all AI is generative; Discriminative AI is the far less talked about alternative, used for classification. This kind of AI can sort existing data (including images, as long as the discriminative AI has computer vision like GPT-4V) into existing classifications (such as pictures of cats vs pictures of dogs). It is essentially assigning binary values, rather than creating anything new, and is less computationally intensive as a result but also, admittedly, less cool.
