Product video is where AI bills hurt: per-second pricing means every re-take is charged again, so the cheapest model per second is not always the cheapest finished clip.
Our pick
Veo when the clip needs native audio — you pay per second and the audio saves a step. Runway when the work needs an editing suite around the model rather than raw generation, at the cost of credit-based pricing that is hard to predict. Open-weights video models when you have the GPU and the volume — free per clip, expensive in hardware and setup.
What one job costslower is better
The 1 cheapest of 1 models with a published price. Bars are logarithmic so the cheap end stays readable; every row carries the exact figure.
Veo 3.1
Google · VideoGenerates native audio alongside video. Billed per second of output, which adds up quickly on iteration.
Quality —
Runway Gen-4
Runway · VideoThe most complete editing suite around the model, not just raw generation. Credit-based pricing is hard to predict month to month.
Quality —
HunyuanVideo
Tencent · VideoThe leading open-weights video model. Needs a serious GPU — this is not a laptop workload.
Quality —
Remove a filter, or submit the model you are looking for.
Prices checked hourly against the LiteLLM dataset · last updated 28 minutes ago. API rates, not seat subscriptions. Watching 320 models for new releases.
What this task actually costs
A 10-second 1080p clip is the working unit here. On a per-second API model it is a fixed number you can multiply by your take count; on credit-based plans, credits-per-second varies by resolution, so the effective rate is what we calculate. Every estimate on this site shows its arithmetic.
The catches, out loud
- Per-second pricing punishes iteration — budget for the takes, not the final cut.
- Max clip length caps at 5–10 seconds on most models; a 30-second ad means stitching clips, which shows.
- Product accuracy is still unreliable: logos, labels and exact colours drift between takes.
See also: all video models · all best-AI guides.