Transcription
Here are six tips to help you write the absolute best prompts in the Sora AI video generator.
One of Sora's strengths is that it's really creative and imaginative. You can combine unique elements together in an unexpected way to get really amazing results. An easy way to visualize this is by generating hybrid animals, like a tiger and eagle combo. I created a bunch of these animal combos, and Sora is able to seamlessly blend the different body parts and patterns together, which really showcases its creativity.
For this horse and leopard hybrid, it mimics the galloping motion of a horse but maps the leopard's patterns onto the skin. Now, if you're using Sora, that means you have access to ChatGPT, which is pretty useful when it comes to writing prompts. As an example, I asked it to write me a short description of a hybrid animal that combines a tiger and an eagle in a cinematic movie, and it came up with a prompt that can be used to generate this impressive shot of a tiger eagle flying in the jungle.
While ChatGPT is a useful tool, you'll still need to edit the prompts to get the best results, but we'll cover that later in the video. Sora is pretty good at generating animals in general, whether that's in the hybrids I showed or anthropomorphic creatures with animal faces but human bodies and actions.
So far, Sora has gotten some mixed reviews, with some people being impressed by the creativity and imagination, and other people pointing out problems with it, which I'll also talk about more later on in the video. But one thing it does really, really well is people. People, especially when it comes to details in the faces or unique clothing with different textures and patterns, and also putting people inside some crazy imaginative environments.
The prompt adherence is amazing. It's able to understand the ethnicity, the braided platinum hair, her cloak made of scales, and her face painted with patterns made from ash. The environment of being on a boat in a lake with glowing creatures in the background are all parts of my prompt that Sora follows correctly.
The videos I'm showing are fantasy-style environments and situations, but you can also just make videos of people doing regular everyday activities as well. However, just like in other video generators, it does make a lot of mistakes when it comes to anatomy. A few extra arms or legs here and there happen a lot, so I avoid generating videos where there's a lot of dynamic movement and motions in them.
If you want the most accurate anatomy, by the way, subscribe if you want to see more prompt guides like this. I've got more on the way.
Tip number three is using different camera shots. This can mean the distance that the video is taken from. You can be really close to the subject and highlight emotions and faces, pull back a bit further to show more context in the scene, and have the characters interact more with the environment. Use wide-angle shots to capture the full environment. These are great for emphasizing the magnitude of everything compared to the size of the character.
Full body shots or long shots also work well to add context and scale to the scene. You can also change the angle that you're viewing the characters from. Low angle shots place a camera below and tilt upwards, making them look more imposing and dominant, like they're in charge. High angle shots move the camera above and point down. These are used to make characters look smaller and more vulnerable and convey a sense of solitude or loneliness.
Bird's eye view flies a camera high into the air, overlooking the entire environment. These are great to use along with establishing shots to set up the location and build the world for your AI films.
So we've seen that camera shots work pretty well in Sora, but one of the problems is that camera movements really don't work out well based on what I've seen so far. By default, Sora already adds in a ton of camera motion, more than other AI video generators. You get a lot of shaky camera work and scene swapping to begin with.
Then, when you try to add your own camera motions on top of that, it just seems to encourage more scene swapping and wild shaky camera work that can be pretty unnecessary. I asked for a figure walking in the woods and the camera to zoom in on their face, and in the video, you'll see someone's face pop up on the screen out of nowhere and suddenly disappear.
When you're prompting, you need to try to avoid anything that can be interpreted as adding in additional scenes or shots. I'm not saying camera motions don't work at all. Here's a good example of panning inside a busy market in Baghdad, but it definitely is one of Sora's weak points.
Also, if you're using ChatGPT to write prompts for you, check them before running them in Sora. If they describe camera motions, which ChatGPT really likes to do when you ask it to describe a video scene for you, you might want to remove those camera motions to avoid any weird stuff going on.
I've also found that when trying to use the image-to-video feature, where I'll attach a reference image for Sora to animate, if I include camera motions in my prompt, Sora almost always seems to do some weird extreme camera work and either changes the scene completely or sways the camera around. Even simple and straightforward prompts and images like this, where I ask for the camera to pull back a bit, it does still add in blurred motion and zooms in on the character.
While camera movements don't really work that well when it comes to your prompts, adding new stylization does work really, really well. GoPros are great for action shots and can be a bit disorienting, while camcorders can make for good horror films.
Revisit different time periods with keywords like "a video from the 1910s" or just prompt for "vintage film," which gives us historical old-school color grading like you're watching in an archive. Color grading can also add in a big difference in terms of the emotions you convey in a video, and Sora is pretty good at following these.
Cool blue dark tones are mysterious and moody, perfect for serious themes. Warm orange and red tones feel cozy, comforting, and are pretty positive overall. Pastel colors with light pinks, greens, and blues are light-hearted and airy, good for cheerful and more relaxed moods. But these saturated colors give the opposite vibe; they provide emotions of gritty dark situations, even conveying hopelessness.
Monochromatic color palettes are also pretty common, where everything follows a single shade, like purple. One thing that did work incredibly well in Sora, more than other AI video generators I've seen, is the lighting.
Lighting is one of the most important elements when it comes to a video's visual style, and most video generators aren't really that good at this. But Sora captures so many different lighting styles, like diffuse and beautiful soft glowing lights and bright backlit lights to really highlight the silhouettes of your characters.
Effects like lens flare work well, perfect for this video in space that's extra exposed to the sun's rays. But you can also dim down the brightness with subtle lighting from a campfire at night. Try different lighting sources, like a lamp on a desk or light shining through windows that capture unique shadows in the subject's face.
There's an unlimited number of colors and light sources you can choose from, and they all work beautifully in Sora. If you're interested in Sora but you want to know how it ranks against the other best AI video generators, go watch this tutorial right here.