These text-to-video APIs were tested with one deliberately awkward brief: a small white paper airplane crossing a bright office while the camera tracks it. The test checks more than a pretty first frame. It asks whether a model can keep a small object recognizable, carry it through a moving shot, hold shallow depth of field, and avoid turning the office into a different room halfway through the clip.
The text-to-video API test
Every run used this prompt: A white paper airplane glides through a sunlit open plan office. Slow tracking shot following the airplane. Shallow depth of field. Realistic. The prompt has one moving foreground subject, repeated office geometry, bright window light, and an explicit camera instruction. That makes it useful for spotting subject drift, inconsistent furniture, and camera moves that stop following the action.
This is a one-output comparison, not a benchmark or a ranking. A single generation cannot establish a model-wide quality score. It does show how each selected configuration interprets the same request, and it records the settings and elapsed time from the runs already embedded below.
What the five outputs actually show
Kling v3
The Kling v3 sample is a 5-second, 16:9, 720p standard-mode run. It took 189 seconds. The important thing to inspect is the relationship between the airplane and the camera: the shot keeps the subject as the visual reason for moving rather than treating the tracking instruction as background decoration. The office detail and focus falloff make the prompt feel photographic, but the short duration leaves little room to judge a longer uninterrupted move.
Kling v3 exposes 5-, 10-, and 15-second durations, 16:9, 9:16, and 1:1 ratios, optional sound, and an optional multi-shot mode. That makes it the broader fit when a draft needs a longer clip, vertical delivery, or several directed beats. See the Kling v3 model page. Kling’s official release notes provide the model-maker reference used here.
Kling v3 Omni
The Omni sample also uses a 5-second, 16:9, 720p standard text-only setup and completed in 201 seconds. In this baseline, it should be judged like Kling v3: does the airplane remain a coherent small object while the office passes behind it, and does the camera read as a follow rather than a generic glide? Its extra reference controls were intentionally not used, so the clip does not demonstrate Omni’s image or video-guided workflows.
That distinction matters. Omni accepts up to seven reference images without a video, or up to four with a reference video; it also supports first and last frames, reference-video modes, sound controls, and multi-shot generation. Pick Kling v3 Omni when an art direction still, a character reference, an end frame, or a reference clip is part of the job. For a pure prompt-only five-second draft, this output is a baseline rather than a reason to assume those controls were tested.
Seedance v1 Pro Fast
Seedance v1 Pro Fast was set to 720p, 16:9, 5 seconds, watermark off, and camera-fixed off. It completed in 54 seconds at 1248×704. That is the clearest speed contrast in this set. The camera was left free on purpose because the request calls for a tracking move; a fixed camera would test a different behavior.
Watch whether the airplane stays legible as it crosses the frame and whether the office retains its open-plan layout once motion begins. Seedance offers 480p, 720p, and 1080p, plus 5- or 10-second output and several aspect ratios. Use Seedance v1 Pro Fast for quick prompt iteration, especially when a team needs to compare camera wording before committing to a slower pass. The speed result here belongs to this 720p, five-second configuration only.
PixVerse Text-to-Video v5
The PixVerse v5 output uses a 5-second, 16:9, 720p run with no style preset, a randomized seed, and watermark disabled. It completed in 43 seconds, the shortest measured elapsed time in this set. The small airplane is the pressure point: a smooth office move is not enough if the object changes shape, scale, or position from frame to frame.
PixVerse offers 360p, 540p, 720p, and 1080p quality choices; 5- and 8-second durations; a negative prompt; style presets; and a seed. The eight-second note in the model documentation carries a resolution limitation, so settings should be checked before relying on it. Choose PixVerse Text-to-Video v5 when turnaround matters and a short social or concept clip is the goal. The PixVerse official site is the model-maker source for the product.
Hailuo 2.3
Hailuo 2.3 ran at 768P for 6 seconds with prompt optimization enabled and completed in 90 seconds at 1366×768. The extra second changes the viewing test slightly: it gives more time to see whether the airplane and the tracking move stay coherent after the opening beat. Prompt optimization was left on, so the result reflects the model’s assisted interpretation rather than strict literal prompt following.
Hailuo 2.3 supports 768P and 1080P. Its documentation limits 1080P to 6 seconds, while 10-second clips are available at 768P. Turn prompt optimization off when wording must be followed tightly, and leave it on when a stronger automatic interpretation is acceptable. Pick Hailuo 2.3 when a 6- or 10-second motion sequence and its prompt-optimization control suit the brief. Hailuo AI is the model-maker source linked for product context.
Parameters, run time, and cost notes
| Model | Run settings | Output | Elapsed time | Cost |
|---|---|---|---|---|
| Kling v3 | std, 5s, 16:9, sound off | 1280×720 | 189s | Not shown in the captured run record |
| Kling v3 Omni | std, text only, 5s, 16:9, sound off | 1280×720 | 201s | Not shown in the captured run record |
| Seedance v1 Pro Fast | 720p, 5s, 16:9, camera-fixed off, watermark off | 1248×704 | 54s | Not shown in the captured run record |
| PixVerse v5 | 720p, 5s, 16:9, no style, random seed, watermark off | 1280×720 | 43s | Not shown in the captured run record |
| Hailuo 2.3 | 768P, 6s, prompt optimizer on | 1366×768 | 90s | Not shown in the captured run record |
No per-output prices are stated here because the available model documentation and the saved runs for these five clips do not expose a comparable cost value. Reporting a guessed rate would make the comparison worse, not better. Times are wall-clock elapsed seconds from these outputs, so queue time, provider load, and a different resolution can change them.
Which text-to-video API should you pick?
Start with PixVerse v5 or Seedance v1 Pro Fast for a fast five-second concept pass. The measured runs finished in 43 and 54 seconds, respectively. Pick Hailuo 2.3 when six seconds at 768P or a longer 768P option better fits the shot, and decide whether automatic prompt optimization helps the brief.
Use Kling v3 when the project needs the wider duration range or multi-shot control. Move to Kling v3 Omni when references matter: a product still, a character image, an intended first or last frame, or a source video changes the task from plain text-to-video into directed generation. The Omni result above does not prove reference quality because it was purposefully run without one.
For more context, compare this test with LTX-Video vs Kling vs Seedance: 5 Text-to-Video Tests, Seedance 2.0 vs Seedance V1 Pro Fast, and Top 5 Image-to-Video APIs in 2026.
Run the same prompt on the five model pages above, then change one variable at a time: duration, ratio, camera wording, or a reference image. That is the quickest way to turn a promising sample into a useful production choice.
