AI video generation sounds like the dream. Type a prompt, hit return, and out comes a short film, right? Not quite. While the technology has taken huge strides, especially in generating still images, video is still very much a work-in-progress — especially when it comes to human beings.

One of the biggest problems? Limbs. Arms, legs, fingers — anything that bends or moves — consistently trip up even the most advanced models. In one frame, the character’s arm might be holding a cup; in the next, it’s fused with their torso or growing out of their back. Consistency across frames is incredibly hard for AI to maintain, because most models are still primarily trained on individual images, not sequences.

This leads to a kind of frame-by-frame amnesia. The AI might generate a beautiful still image, but when you stitch 24 of those together per second, the inconsistencies stack up fast. Joints warp, heads twist too far, and fingers morph into awkward blobs. The results can be surreal — sometimes funny, sometimes unsettling, rarely usable without heavy editing.

The issue goes beyond just limbs. Lighting changes unexpectedly, clothing folds in physically impossible ways, and background elements flicker or disappear. And while some tools allow for motion tracking or reference video, it’s still no guarantee that the AI will maintain spatial logic from frame to frame.

Why is it so hard? Because video requires understanding not just space, but time. Motion needs to be smooth and coherent. And humans — with all our subtle gestures, asymmetrical features, and complex movements — are particularly difficult to simulate. Even small inconsistencies break the illusion fast.

That doesn’t mean AI video tools are useless. For abstract visuals, fast concepting, or heavily stylized looks, they can be incredibly powerful. But for anything involving realistic human motion, they still need a lot of help. Often, the best results come from combining AI generation with manual post-production — frame edits, rotoscoping, or even full animation passes to clean things up.

In short: AI video generation is impressive, but it’s not magic. It’s a tool, and like any tool, it has limits. The tech will continue to improve, but for now, expect to spend as much time fixing as creating — especially if your video includes people doing anything more complex than standing still.