Kling 3.0 Image-to-Video: Visual Notes on Camera Movement Language

Starting from a fixed-camera portrait reference frame as the first frame, the Kling 3.0 image-to-video pipeline breaks common camera moves into reproducible visual demonstrations: beginning from the first frame, the model completes push-in, pull-out, pan, and tracking trajectories within the same subject and scene, for use in storyboard previsualization, cinematography teaching, and comparing movement plans before a shoot. In terms of composition, the work relies on only one media input: the first-frame image. It is generated by the pipeline and bound as image_urls[0]; the schema requires only image_urls as a mandatory field, so base carries only non-media parameters—a default 5-second duration, 1080p resolution, and native audio disabled—making it easy to first verify at a low-cost tier whether the camera move works, then extend the duration or raise the resolution as needed. The prompt is used to guide movement direction and pacing, not to replace the reference frame. Use cases: in film and advertising pre-production, writing "how this shot moves" as a visual specification; short-video teams building an internal reference library of camera moves; developers debugging the relationship between first-frame composition and motion control in image-to-video. Note that the single-run duration of 3–15 seconds still requires segmented generation and post-production stitching for complex long takes, and camera-move stability should be judged by the actual output.