The limits, stated

• 3–8 seconds per generated shot, at 720p portrait, 24 fps. A motion shot is one clip; longer scenes are cut from several, or held on a still. Hero quality finishes at 1080p portrait.
• Motion starts from a still we render first. You approve the picture, then it moves — measured keyframe fidelity SSIM 0.912, over six frozen test frames. It is not a text-to-video guess.
• Speech is generated inside the clip in Urdu, Hindi, Arabic and English, checked by transcribing the clips back rather than by ear.
• What stays still: clocks, dials, gauges, charts, dense text, handwriting. The model repaints them wrong, so scenes whose meaning is in an instrument keep their still image.
• Motion is rationed to 3 shots a video outside the motion-native formats: at a measured 10.6 seconds of render time per second of video it is the most expensive thing the pipeline draws.
• Long-form stays on stills. Five to twelve minutes of narration over painted frames, exactly as before — and every board you already own still renders and still opens in the same editor.
• Delivered at 720p or 1080p, 9:16 or 16:9 by format, with the title, description, tags and captions written. There is no 4K path and no publishing integration.

Motion video is licensed for commercial use.