Stability AI’s new virtual camera turns any image into a cool 3D video and I’m blown away by how good it is

Stability AI's videos have infused text and images with movement and life for a few years but are now literally adding a new dimension by turning two-dimensional images into three-dimensional videos.

The company's new Stable Virtual Camera tool is designed to process even a single image into a moving, multi-perspective video, meaning you could rotate around and view the film from any angle.

It's not entirely a new concept, as virtual cameras have long been a staple of filmmaking and animation, letting creators navigate and manipulate digital scenes. But Stability AI has taken that concept and thrown in a heavy dose of generative AI. The result means that instead of requiring detailed 3D scene reconstructions or painstakingly calibrated camera settings, Stable Virtual Camera lets users generate smooth, depth-accurate 3D motion from even a single image, all with minimal effort.

What makes this different from other AI-generated video tools is that it doesn’t just guess its way through animation and rely on huge datasets or frame-by-frame reconstructions. Stable Virtual Camera uses a multi-view diffusion process to generate new angles based on the provided image so that the result looks like a model that could actually exist in the real world.

The tool lets users control camera trajectories with cinematic precision, choosing from movements like zoom, rotating orbit, or even a spiral. The resulting video can be in vertical form for mobile devices or widescreen if you have a cinema. The virtual camera can work with just one image but will handle up to 32.

Stability AI has made the model available under a Non-Commercial License for research purposes. That means you can play with it if you have some technical ability by grabbing the code from GitHub. Going open-source as Stability AI usually does also means the AI developer community can refine and expand the virtual camera's capabilities without the company needing to pay.

3D AI

Of course, no AI model is perfect, and Stability AI is upfront about the kinks still being worked out. If you were hoping to generate realistic people, animals, or particularly chaotic textures (like water), you might end up with something that belongs in a low-budget horror film.

Don't be surprised if you see videos made with it featuring perspectives that awkwardly travel through objects or have perspective shifts leading to flickering, ghostly artifacts. Whether this will be a widely adopted tool or just another AI gimmick ignored by dedicated filmmakers remains to be seen.

Not to mention how much competition it faces among AI video tools OpenAI's Sora, Pika, Runway, Pollo, and Luma Labs' Dream Machine. Stable Virtual Camera will have to show it performs well in the real world of filmmaking to go beyond just another fun demo video.

You might also like

• I tried adding audio to videos in Dream Machine, and Sora's silence sounds deafening in comparison
• YouTube Shorts might soon be filled with AI videos
• TikTok owner ByteDance has a new AI video creator you have to see to believe