Luma AI just shipped Ray3.2, the latest update to its Ray3 video generation model, and the headline feature is something the AI video space has been badly missing: real directorial control. You can now set up to 16 keyframes inside a single clip, telling the model exactly what changes, what holds, and when the story lands. That is a meaningful jump from the start-and-end-frame approach that defined the previous generation.

The Control Problem, Finally Addressed

Ray3 Modify was built to address one of the fundamental limitations of early AI video systems: their inability to reliably follow and preserve human performance. Historically, it has been difficult to preserve timing, motion, and emotional intent during acting and scene transformations. Ray3.2 pushes that further. Ray3.2 transforms creative intent into scalable video workflows with richer control, continuity, and cinematic direction.

The model ships three major capability pillars:

• Multi-Keyframe (up to 16): Direct what changes frame by frame inside a single clip. Set the shot, compose the story, and let the model interpolate with cinematic fidelity.
• Modify Video V2: Drastically improved and now better maintains the physical logic, narrative coherence, and performance authenticity of your original footage. It supports layered, scene-aware transformations that behave as though they were captured naturally in camera. Clips up to 20 seconds, all at 1080p.
• Reframe: Adapts any clip to any aspect ratio. Supports 6 aspect ratios: 9:16 (portrait), 3:4, 1:1 (square), 4:3, 16:9 (landscape), and 21:9 (ultrawide).

What the Model Actually Does Well

The Ray3 family has been building toward a specific niche: hybrid-AI production, where a real camera and real actor provide the raw material and the model handles the transformation. With Ray3 Modify, the human performer, camera operator, or physical input becomes the source of direction for AI. This enables AI to follow real-world motion, timing, framing, and emotional delivery. By conditioning AI generation on human-led input footage, Ray3 Modify significantly reduces guesswork and enables creators to guide shots closer to their original intent.