# Seedance, Hailuo and Kling: evidence scope

The article compares three explicitly named API variants: Seedance 2.0 Fast image-to-video, Hailuo 2.3 Standard image-to-video and Kling v3 Pro image-to-video. It is not a survey of every current model release.

Two outputs were actually generated using the same unchanged public-domain Lake McDonald JPEG and the same motion prompt. Seedance was generated September 16, 2026 and is reused from `/blog/assets/photo-to-video/output.mp4`; Kling was generated September 19. Hailuo's entry is documentation-only; no Hailuo sample or performance score is presented.

Source image: `/blog/assets/photo-to-video/source.jpg`, Ryan Hagerty/USFWS, Public Domain, https://www.fws.gov/media/lake-mcdonald-mt . The source agency did not produce or endorse the animations.

Seedance request, result and measured metadata are in `/blog/assets/photo-to-video/`. Kling request, selected actual result fields and ffprobe metadata are in this directory. Both requests target five seconds and disable audio. The native output dimensions differ; this is not a resolution-controlled benchmark. No latency comparison was measured, and one sample per model cannot establish a general quality winner.

Observed generation charges: Seedance 135 cents; Kling 185 cents. These are historical charges for the recorded calls, not a standing quote. An initial Kling request failed input validation and charged zero cents; the successful request is preserved here.

The worksheet leaves unmeasured quality scores empty. Use repeated matched trials on your own task before making quality rankings. Keep failed runs and rerun costs in the record.