I'm showing you how to create the most cinematic AI videos using Runway gen 3 when it comes to making the best AI videos a big part of that is writing the right prompts that's something I struggled with a lot when Gen 3 first came out and I wasn't happy with the results I was getting but after testing it out I have some strategies to share with you on how to write the best prompts for cinematic AI videos and make it behave the way you want it to the first thing we'll do is use an image
to video tool to upload our own reference images using reference images is the best way to get high quality results I'll attach a photo of a woman jumping in midair and we'll put it as the last frame of the video this means the AI fills in all the parts before this photo and it'll end the final video frame with this image to start off let's use a simple prompt out write the woman floats in the air let's see the results we get the first major problem gen 3 has is that it tends to change the
color saturation in the reference photo on the left the colors are muted and light but when it's animated the colors are way more saturated at the start of the video the blue sky is a lot more Vivid and there's more contrast to deal with this specify in the prompt exactly the type of color palette you want I'll add in D saturated pastel colors do you see the difference in the generated videos the one with the updated prompt keeps a consistency in the color palette all the way through let's see another example I'm starting with this
reference image that has the these muted pastel pinks in the prompt on the left I only prompt for the motion of looking around with a camera panning and the colors slowly but surely become more saturated with higher contrast her face also changes to a older lady on the right I add in desaturated colors to the prompt and look how consistent the colors stay throughout the entire video her face also doesn't change other keywords you can use to preserve the color palette include muted colors and low contrast so far we're off to a good start now
now it's going to get a little trickier camera movements are really hard to control in Runway gen 3 for some reason it likes to zoom in a lot even when I specifically prom for it to zoom out or for the camera to pull back it still Zooms in on a lot of landscape images like this one despite this we can still try and make the best out of the prompt motion control try using static shot for reduced camera motion think about exactly how the video is being taken I prompt for a drone shot flying high
into the air for a bird's eyee View and the camera elevates into the air revealing other structures behind the temple adding other environmental Clues to guide the movement in this prompt the camera tilts up to the moon in the sky since the moon's above the temple it helps a camera motion tilt upwards one keyword that works really well is fpv fly through which stands for firstperson View and the camera flies forward and gives something like a drone video instead of only adding in camera keywords think about how the addition context you provide in the prompt
will affect the camera motion let's animate this side profile of a woman I asked for the camera to zoom in on her eyes asking for the eyes tells gen 3 to keep the focus on the face although the zooming effect didn't work as well as I'd hoped asking to see the eyes also makes her turn her head to face us I can also prompt for a zoom out with a full body shot of a black dress because I mentioned her clothing it helps gen 3 zoom out more to capture more of the body however I
want to keep her face inside the shot so I'll add as she smiles to that prompt now the zoom out effect is much more subtle but her face stays inside the video when you use reference images you can also use the reference as the last frame of the video to Great effects after uploading an image reference make sure the last frame option is checked add a prompt that describes a motion that would roughly end up with this image as the last frame by the way this doesn't work in turbo mode right now it's only in
gen 3 Alpha it's a good way to control the movement since you know exactly where the video ends up there are some limits to its prompt understanding though it won't always generate exactly the motion you asked for although some of the videos can turn out really well an important part of getting the right prompts is using an iterative process and refining your prompts slowly over time you'll have to try out different keywords and move them around let's walk through the process so if I want to animate a photo of this person and and have her
talking and smiling at the same time using what we've learned earlier I'll start with muted colors low contrast to preserve the color palette and then add the woman Smiles as she's speaking the motion in the video looks good but you'll notice two Eiffel Towers appearing in the video that's obviously AI generated so I adding static camera into the prompt so there's less movement which prevents that extra Tower from showing up however using the word static limits the motion overall now now the woman only smiles and doesn't speak the next thing I tried is moving the
phrase smiling and speaking to the start of the prompt so that it's weighed more heavily this kind of works you consider mouthing a few more words now but it's not really the dynamic speaking motion I wanted as it shot in the dark I replace the word speaking with talking and there it is the background stays consistent without an extra Tower warping in and she's speaking with a faint smile sometimes AI responds in unexpected ways to different keywords prompt ing in Runway is an iterative process and can't get frustrating if you don't have the right tools
and strategies I hope this video gave you some better ideas for how to get the right prompts if you want to learn how to generate AI visual effects using Runway gen 3 go and check out this video tutorial I made over here