Video render time
How long each video model takes measured on OceanToken
179 timed clips across 7 video models, October 15, 2026 to November 12, 2026. A setting is shown only once it has at least 5 timed clips.
Quickest clip
51 s
Veo 3.1 Lite
6 s clip, 720p
Typical clip
173 s
Median across the 17 settings below
Upscaled vs native
+60 s longer to render, typically
Price per second 19% lower
Upscaled or native
An upscaled model renders the clip at a lower resolution and enlarges it. At normal viewing size, on a phone or in a web player, the two are hard to tell apart; zoomed in to the pixel, fast motion, skin and hair, and fine texture can look smeared or over-sharpened. Composition, motion, finger counts and spelling come from the base model, not from the upscaling.
| Model | Setting | Median time | Frame delivered | From | |||
|---|---|---|---|---|---|---|---|
| Native | Upscaled | Native | Upscaled | Native | Upscaled | ||
| Seedance 2.0 | 5 s clip, 720p | 181 s | 219 s | 1280×720 | 1256×720not the standard frame | $0.14/s | $0.11/s |
| Seedance 2.0 Fast | 5 s clip, 720p | 115 s | 197 s | – | 1256×720not the standard frame | $0.098/s | $0.085/s |
Every setting measured
Quickest first. Time runs from submitting the job to the finished clip being ready to download.
| Model | Setting | Median time | 9 in 10 within | Frame delivered | From | Clips |
|---|---|---|---|---|---|---|
| Veo 3.1 Lite | 6 s clip, 720p | 51 s | needs 10 clips | – | $0.027/s | 5 |
| Grok Imagine Video | 5 s clip, 720p | 66 s | needs 10 clips | – | $0.070/s | 5 |
| Seedance 2.0 MiniCompare | 4 s clip, 480p | 99 s | 134 s | 864×496 | $0.028/s | 42 |
| Seedance 2.0 Fast | 4 s clip, 480p | 100 s | 128 s | 864×496 | $0.045/s | 14 |
| Seedance 2.0 Fast | 5 s clip, 720p | 115 s | needs 10 clips | – | $0.098/s | 5 |
| Seedance 2.0 MiniCompare | 5 s clip, 720p | 121 s | needs 10 clips | – | $0.061/s | 5 |
| Kling v3.0 Standard | 5 s clip, 720p | 122 s | needs 10 clips | – | $0.080/s | 5 |
| Seedance 2.0 Mini UpscaledCompare | 4 s clip, 720p | 166 s | 180 s | 1256×720not the standard frame | $0.053/s | 15 |
| Seedance 2.5Compare | 4 s clip, 480p | 173 s | 298 s | 854×480 | $0.10/s | 13 |
| Seedance 2.0Compare | 5 s clip, 720p | 181 s | 252 s | 1280×720 | $0.14/s | 12 |
| Seedance 2.0 Fast Upscaled | 5 s clip, 720p | 197 s | needs 10 clips | 1256×720not the standard frame | $0.085/s | 6 |
| Seedance 2.0Compare | 4 s clip, 480p | 200 s | 271 s | 864×496 | $0.067/s | 12 |
| Seedance 2.0 UpscaledCompare | 4 s clip, 720p | 211 s | 278 s | 1280×720 | $0.11/s | 12 |
| Seedance 2.0 UpscaledCompare | 5 s clip, 720p | 219 s | needs 10 clips | 1256×720not the standard frame | $0.11/s | 7 |
| Seedance 2.5 UpscaledCompare | 4 s clip, 720p | 239 s | 336 s | 1280×720 | $0.19/s | 11 |
| Seedance 2.0 Upscaled 4kCompare | 4 s clip, 4K | 246 s | needs 10 clips | 3840×2160 | $0.55/s | 5 |
| Seedance 2.5 UpscaledCompare | 5 s clip, 720p | 248 s | needs 10 clips | 1278×720not the standard frame | $0.19/s | 5 |
On customer traffic, last 30 days
Every video job our customers sent, counted only where a setting had at least 30 jobs from at least 5 different customers. Refused means the model maker’s content review declined the prompt or the clip.
| Model | Setting | Success | Refused | Median time | 9 in 10 within | Frame as asked | Jobs |
|---|---|---|---|---|---|---|---|
| Seedance 2.0 Mini | 480p, up to 5 s | 100% | 0% | 122 s | 141 s | – | 77 |
| Seedance 2.0 Mini | 480p, 6–10 s | 100% | 0% | 144 s | 173 s | – | 52 |
Time, not looks
This page answers how fast and at what frame a model delivers. For which clips people prefer, see the human-preference votes of Artificial Analysis Video Arena.
How we measure
These are our own test jobs through the OceanToken API, not customer traffic: text-to-video prompts at the resolution and length shown, 16:9 unless stated. Each job is timed from submitting it to the first status check that reports the finished clip (we check every 5 seconds), so queueing at the model maker counts. Failed and refused jobs are not timed. A setting needs 5 timed clips to be shown, and the “9 in 10” figure needs 10. Output frames are read from the delivered files. Render time changes with the time of day and the maker’s load, so yours may differ.