We used one input image and one motion prompt across 3 runs to compare video generation for a person fanning with a uchiwa. Veo 3.1 Lite was usable from the first generation (Pass@1: Yes), all 3 runs were usable, and its cost per usable output was the lowest at $0.12. Veo 3.1 and Kling 3.0 V3 Standard only produced a usable output within 3 attempts. Seedance 2.0 was rejected before generation due to a content policy violation on the person image.
Results
| Model | Pass@1 | Pass@3 | Usable Rate | Avg Score | Total Cost | Cost / Usable |
|---|---|---|---|---|---|---|
| Veo 3.1 Lite | Yes | Yes | 100% | 3.87 | $0.3600 | $0.1200 |
| Veo 3.1 | No | Yes | 33% | 3.63 | $2.4000 | $2.4000 |
| Kling 3.0 V3 Standard | No | Yes | 33% | 3.73 | $1.0920 | $1.0920 |
| Seedance 2.0 | — | — | — | — | $1.2136 (estimated) | — |
Key takeaways
- Veo 3.1 Lite was the only model with Pass@1 = Yes and 100% usable rate in this single-prompt, single-image benchmark.
- Total cost for Veo 3.1 Lite was $0.36 across 3 runs, giving a Cost / Usable of $0.12.
- Veo 3.1 and Kling 3.0 V3 Standard eventually produced a usable output, but cost per usable output was much higher.
- Seedance 2.0 rejected the request because the input image contained a person; this model is not suitable for this subject.
Verdict
For I2V of a person fanning with a uchiwa, Veo 3.1 Lite is the clear winner on first-try success, consistency and cost in this limited test.
Note
This is an English summary of the full Japanese benchmark article. Methodology, prompts, all outputs and cost details are documented in the original post.
Read the full article in Japanese →