Removing an object from a video is easy with AI. Removing it without destroying everything behind it will leave VFX artists pulling their hair out trying to fix the mess AI created.
Take the example in this video. Two people are walking their bikes through a shot. We remove the bikes, but that leaves us with another challenge: recreating everything they were blocking, including parts of the people and the background. And because the people are moving, those reconstructed details have to look natural and stay consistent across every frame. Otherwise, you end up with distorted images, flickering, and more manual cleanup for the VFX artist.
That's exactly what we focused on when building the cleanup tool in StyleFrame. With it, you can:
(1) Remove multiple people or objects from the same shot.
(2) Separate foreground and background elements so you can work on them independently.
(3) Reconstruct missing areas between overlapping subjects, rather than just filling in the space around them.
(4) Preserve the video's original aspect ratio, duration, and timing, frame by frame.
Removing an object shouldn't leave artists with hours of extra work. We want VFX artists to spend less time fixing distorted backgrounds and flickering frames, and more time getting the shot right.
A typical shot can involve 3 to 5 different jobs in a traditional VFX pipeline. With StyleFrame, one artist can handle the entire process.
This is a detailed example we created using footage of an actor against a green screen. Instead of our tool treating the entire video as a single pass, we designed the workflow around the way VFX artists have worked for decades:
Separate the shot into layers.
Make the necessary changes to each layer.
Composite everything back together into the final shot.
For context, whenever we develop a new tool, we start by looking at how professionals already accomplish the task. Then we figure out where AI can take on some of the work without taking creative control away from the artist.
For video editors and smaller creative teams that don't always have dedicated VFX resources, we're excited about what this makes possible.
And for experienced VFX artists, we're building tools that fit the workflows they've spent years mastering.
Much more to come!
my agent made 127 searches and pulled 461 posts just to read my own 105
there's no "give me all of it" button for agents.. so they brute force the human version
agents are going to search 1000x more than us and most products are still built for one person scrolling
Introducing Griffin, the first model to pass the video Turing test.
48% of people who talked to it live thought it was a real human. Previous systems have had a pass rate <3%. It is #1 on NVIDIA's benchmark for full-duplex AI video.
It’s the first Human Interaction Model (HIM).