Pika.art, an AI video creation site
Pika is a platform for creating videos using AI technology and is available for discord and sites. I will explain how to create an AI video through the Pika.art site. You can create a simple AI video based on the prompt entered by the user, and you can enter not only text but also images or videos as parameters, so you have more freedom to create the desired video, and above all, it is much simpler than other methods.
Creating AI videos with Pika.art
First, the main screen of Pika. There's an Explore tab to browse AI videos created by others, a My library tab to view your own creations, and a prompt at the bottom to create a video.
I decided to bring Michelangelo's Creation of Heaven and Earth to life, and I thought it would be a good idea to create a video that gives the impression of something floating, and of two people getting closer and closer together. To start, I downloaded a copyright-free image of the Creation of Heaven and Earth, and uploaded it by clicking the Upload Attachment button at the bottom of the prompt input UI.
Next, type in the appropriate prompt to describe the scene you want your video to represent. Based on other people's videos and reviews, it seems to work better with sentences rather than words. Since there is a base image, you can skip the appearance description and simply enter the prompt. (For reference, the maximum length of the prompt is about 500 characters.)
After a few moments, the My library tab updates with the AI film I created. The wind-blown hair and beard are impressive, and it looks like my fingers are getting closer and closer, albeit subtly. There's a bit of uncanny valley in the head movement, but I'm really pleased with the quality of the video for something so simple.
How to get more out of Pika
Like other image creation tools, pika offers a few more features that allow you to control the outcome of your video. You can adjust the aspect ratio, control camera movement, and negative prompts, all of which can be set on the right side of the prompt input window. You can also add an extra 4 seconds to an already created video, which is a great way to lengthen your footage while maintaining consistency.
Pika's pricing works like this: 10 credits for every 3 seconds of video you create, and 30 credits every 30 days. Paid plans allow for more credits and advanced features like upscaling and watermark removal.
How does Pika work? Different ways to create AI videos
To find out what technology Pika uses, you have to sign an NDA (non-disclosure agreement), which means they won't tell you. According to their description, they have a proprietary AI video creation technology that is not publicly available.
Pika is not the only way to create AI images. Image generator, controlnet, etc. can be used to generate continuous motion frame by frame. There are also many attempts to create V2V videos with Animatediff, which extracts images and control nets from the original video. I'm guessing that Anyone Animate, which I introduced earlier, is also an application of this method.
Another approach is from the "Drag Your Gan" paper. My guess is that Pika.art or Meta's Emu went in this direction. It's characterised by higher consistency than stable diffusion, but I'm not sure how it works. If you're interested, it's worth doing some research on different ways of generating AI video and trying it out.
