Reddit proposes measuring video models by usable seconds, not render time
Good-Razzmatazz-6179 · reddit · 2026-07-28
A Reddit user argues that raw generation time is a misleading metric for video models and proposes tracking minutes per usable final second instead.
Their reasoning is simple: a model that renders in 4 minutes but needs 8 rerolls is slower in real production than one that takes 12 minutes but lands on the second try. The metric should therefore include the full wall-clock cost: first generation, dead seeds, prompt rewrites, and interpolation passes.
They say the columns they track are:
- model and quantization,
- total wall time,
- number of clips generated,
- seconds actually kept,
- whether upscaling happened outside the main loop.
Their ideal comparison would be the same footage, same operator, and side-by-side output across models—because screenshotting generation times alone hides most of the real cost.
More from Multimodal
- Third-party test: Claude Opus 5.5 renders finer 3D scenes but costs 13x more than GPT-6 Sol — testingcatalog · 2026-09-23
- ComfyUI trick: aux preprocessor + Qwen transfers poses across characters with one prompt — Acceptable-Work8202 · 2026-09-23
- Same portrait prompt across Midjourney V6.1, V7 and V8.2: do older models look better? — tisch_eins · 2026-09-23
- Testing AI character consistency across a 20-image travel sequence — SiennaVaire · 2026-09-23
- Midjourney v8.2 Faces: New Portrait Generation Samples Shared — azed_ai · 2026-09-23
- One-sentence prompt generates lifelike dog video, shown side-by-side with the real one — wgrathwohl · 2026-09-23