The AI image generator built into chatbots may have attracted most attention these days, but the dedicated AI image engine, Midjourney, has quietly improved and evolved since its launch three years ago. It also has a video model now.
According to Midjourney, this is another step in creating AI tools that can produce real-time 3D world simulators. The V1 model has been released with its ultimate goal in mind, but it will take some time to get there.
Midjourney's AI video maker works a little differently than other generators. You start with an AI-generated image or an image you already have, and MidJourney creates a 5-second animation. These short clips can be extended for a total of 4 times, for 4 seconds at a time.
As always with Midjourney, creating this content takes time (credits for the Midjourney version): The second video costs the same as image generation, with the Midjourney plan starting at $10 a month and rising from there.
Create a video with Midjourney
To create a video in Midjourney, you first need to create an image through the web interface. Use the slider button on the right to set some options, such as aspect ratio, and enter the prompt in the box at the top. As accurate as you can at the prompt (there are more hints here), and hit input (Or click the submit icon) to run it.
As always, Midjourney presents some results of your prompts and offers options to build them. These include four animation options for creating videos. Your first decision is whether to go together Automatic (Midjourney selects the added movement) or manual (We'll explain what you want to do).
The second decision is whether to go together Low motion (Motion is limited) or High movement (where everything in the frame moves and there is a high chance of glitching). Once you have created the pick, you can edit the prompt again (if you select Manual), and the video will be created. As with the image, multiple variations are displayed.
Click on the generated video and you will see that the same four animation options are here. Only these can be run in total four times to further expand the video. You can mix automatic and manual sections, as well as low and high motion sections to build the clip you're looking for.
You will have the option to download the video above the prompt on the right. You can download RAW videos or versions optimized for social media (to combat some of the compression that occurs when you post videos to these platforms). You can start again by clicking on the original prompt and making changes.
Midjourney is an impressive AI image generator, and its videos reach the same standard. I tried to create an animation of the SF Cityscape and natural landscapes, but the end result was almost consistent and logical, but closely followed the prompt instructions. Some of the typical quirks of AI-generated videos are here like strange physics, but even at this early stage, the V1 model is polished and capable.
In these clips you can see both the limitations and benefits of the mid-journey approach. Each 4-second segment moves smoothly to the next segment, but if you're working with a 4-second burst, you don't have much time to do what you want to do with the video.
What do you think so far?
Compare Mid Journey with Sora and Gemini
If you're paying more than $20 a month for ChatGpt, you can also access SORA. Like Midjourney, SORA allows you to start videos from images (either AI generation or other).
I started building SORA on top of the futuristic sci-fi cities and animated landscape images I created with Midjourney, and I got a variety of results. The scenes found it more appealing, but there were some more oddities, such as the unnatural movements and the glitchy background (it got really weird, especially in animation).
You can use SORA to generate videos up to 20 seconds long, but you don't have much more control over how the scene progresses than Midjourney. Basically you enter the prompt and then get what's coming back. At least on casual projects, Midjourney feels like a more accessible tool and gives you more realistic results.
I also tried to create the same scene in Google's VEO 2 via the Flow Online app. Flow lets you create videos based on images and enhanced scenes while still maintaining consistency like Midjourney (I can't get the same functionality in VEO 2 in the Gemini app). Overall, I think this gave me the closest result to what I was looking for, but there was still some contradictions and oddity.
Quick instructions are closely followed by flying cars that come down in a trustworthy way through the cityscape. As for animation, the results for Google Flow and Veo 2 were the best as they fly across the cartoon landscape, but we can see that they gradually lose some of the richness and details that exist in the original image.
If AI's filmmaking ambitions are a little more epic, then Google's tools might be the best, but they're costly. Video generation and access to Flow will bring you back over $20 a month. You can also pay $250 for the Google AI Ultra plan. This also allows you to extend access to more advanced VEO 3 models with sound (VEO 3 still can't create videos based on static images).
Although this is not the largest sample size, the quality of Midjourney Clips is clear and the approach to video creation is simple and intuitive. Google Veo 2 is still a good option for the overall quality, but Sora is pretty confusing and unpredictable for now. Spending more time on the Openai model will give you the possible results.
