First-Party Video Model Benchmark

Uniform prompt and reference input

Every candidate model receives this exact prompt and the same static product-photo reference file. The reference asset is selected from a fixed set before the run and its source URL and SHA-256 hash are recorded with the results.

A red ceramic coffee cup on a pale oak table beside a folded linen napkin. Keep the cup, logo, and handle shape stable while soft morning light crosses the table. Slow dolly-in, natural shallow depth of field.

Evaluation method

• Use one fixed prompt for every model; do not translate, shorten, or optimize it per model.
• Use the same product photo file as the reference for every model, without changing its crop, dimensions, compression, or filename.
• Run two unattended generations per model at 4 seconds and 720p, recording every successful and failed attempt.
• Record client-measured generation time from submit until the terminal success or failure event.
• Record the actual credits charged by imgmov, not only the model's advertised rate.
• Compare output against product preservation, prompt adherence, motion stability, file integrity, and visible artifacts. Use qualitative observations, not invented numeric scores.
• Publish a row only when the run, output URL, settings, cost, and evaluator note are available. Omit failed or unavailable fields as "not recorded".

Observations describe one evaluator's review of these runs. They are context-specific, not universal model rankings or a substitute for a viewer's own test.

Result table

Candidate models and published rates

These rates are imgmov's published list rates, not benchmark costs. The result table will record the actual charge for each completed or failed run.