AI camera moves tested on 8 video models
We gave 8 current AI video models the same prompt for each of 16 camera moves, from a simple dolly in to a dolly zoom, and watched what each one actually did. Every clip below is real model output, one take each.
Scorecard: how often each model followed the move
| Model | Followed | Partly | Missed | Unclear | Cost per clip | Output |
|---|---|---|---|---|---|---|
| Gemini Omni Flash | 6 | 4 | 0 | 6 | $0.50 | 1280×720, 5 s |
| Kling 4 | 4 | 5 | 3 | 4 | $0.64 | 1280×720, 5 s |
| Wan 3 | 3 | 5 | 2 | 6 | $0.50 | 1280×720, 5 s |
| Hailuo 3 | 2 | 6 | 4 | 4 | $0.50 | 1344×768, 5 s |
| Seedance 2 | 2 | 3 | 3 | 8 | $1.50 | 1280×720, 5 s |
| Google Veo 3.1 | 1 | 5 | 4 | 6 | $1.20 | 1280×720, 6 s |
| Runway Gen-4.5 | 1 | 3 | 5 | 7 | $0.60 | 1280×720, 5 s |
| Grok Imagine 1.5 | 0 | 5 | 8 | 3 | $0.80 | 1280×720, 5 s |
Out of 16 moves. "Unclear" means our two reviewers disagreed. Verdicts judge the camera move only, from 8 frames per clip, not image quality. One take per model can't show consistency.
How we tested
Each prompt is word for word what Kingy's AI Camera Simulator writes for that move, on one shared scene: a person in a mustard jacket at a blue-hour coastal overlook, medium shot, eye level, 50mm, 16:9, about 5 seconds. Every model got the same text and no starting image. We ran each once through the Runway developer API on November 7, 2026 and kept every result, including the bad ones.
Clips are shown at each model's own output size: 1280×720 for most, 1344×768 for Hailuo 3. They are not upscaled. Veo 3.1 only offers 4, 6 or 8 seconds, so its clips are 6 seconds; all others are 5. Costs are Runway's API credit charges at $0.01 per credit, no audio.
Two reviewers judged each clip separately from 8 evenly spaced frames, without seeing each other's answers. We publish a verdict only where both agreed; otherwise it says "Unclear". Still frames can't always tell an optical zoom from a camera moving forward, and they don't show smoothness, so treat the verdicts as a guide and watch the clip.
All clips are AI-generated test footage, not real filming. Generated with the Runway API.
Questions
Which AI video model followed camera moves best?
In this test, Gemini Omni Flash followed 6 of 16 moves fully and 4 partly. That's one take per model on one scene, so treat it as a guide, not a ranking.
Which camera moves are hardest for AI video?
No model fully followed 6 of the 16 moves: Crane up, Handheld, Pedestal down, Whip pan, Dolly zoom and Crash zoom. Pedestal down and Dolly zoom were missed most (5 of 8 models each).
How much did it cost?
$99.84 for all 128 clips at Runway's API prices on November 7, 2026. Per clip: Gemini Omni Flash was cheapest at $0.50; prices for each model are in the scorecard.
Are these clips upscaled or edited?
No. Each clip is exactly what the model returned, at its own output size, with no upscaling, trimming or colour changes.