AI video model comparison

How to read this AI video model comparison

Length is the range of clip durations a model accepts in one request. Resolution runs from the smallest to the largest output the catalogue lists, and aspect ratios are the frame shapes you can ask for, from widescreen 16:9 to vertical 9:16. A red tick under sound means the model can generate audio with the picture; a dash means it returns silent video. Hover the frame shapes to read the exact ratios.

Start and end frame shows whether you can pin the opening image, the closing image, or both, with pictures of your own. Price per second is the list price at 720p where a model offers it (otherwise its lowest listed resolution), with sound when the model makes sound. Models billed by video tokens are converted at their 16:9 frame size and 24 frames per second and marked ≈. “Listed” is the date a model appeared in the catalogue, not its release date.

What the AI video model comparison shows today

19 of the 33 models generate sound and 19 let you set both the first and the last frame. 6 render up to 4K, and the longest single clip is 30 seconds (HeyGen Video).

At 720p the lowest list price is $0.010 per second (mischief 1), so a five-second clip costs about $0.05. The cheapest model that also makes sound is Veo 3.1 Lite at $0.050 per second. At the other end, Veo 3.1 lists $0.400 per second at 720p.

Choosing from the comparison

For short vertical clips, filter for 9:16 support and sort by price: a cheap model at 480p or 720p is usually enough for a feed. For a hero shot, look at 1080p or 4K output and native sound, and accept a higher price per second. When a character or product has to stay consistent, favour models with start and end frames, because pinning both ends removes most of the drift between takes. For long single takes, check the top of the length column before anything else.

A spec sheet narrows the field; it does not pick the winner. Once two or three models pass on length, resolution, sound and price, write one prompt that describes the shot you actually need and run it on each of them with the same length and aspect ratio. Look at motion first, then faces and hands, then whether the camera did what you asked. Keep the prompt identical so the difference you see comes from the model, not the wording, and only then tune the prompt for the model you keep.

Prompting matters as much as the spec sheet. The prompt guides cover how each family reads a prompt: veo 3 prompt guide, wan 2.2 prompts, kling ai prompts, nano banana prompts and flux prompt guide.

AI video model comparison FAQ

Which AI video model is the cheapest?

In this AI video model comparison the lowest list price at 720p is mischief 1 at $0.010 per second; with generated sound, Veo 3.1 Lite at $0.050 per second. Prices change, so sort the table by price for today's order.

Which AI video models generate sound?

19 of the 33 models in the comparison can generate audio with the picture. They are the rows with a red tick in the sound column; the rest return silent video that you score separately.

What does start and end frame mean in an AI video model comparison?

It is whether you can give the model your own image for the first frame, the last frame or both. Pinning both ends is the most reliable way to keep a product or character consistent and to control where a shot lands.

Why do prices here differ from the maker’s own app?

The comparison uses list prices from one public API catalogue so that every model is priced the same way. Makers and other platforms sell the same models through subscriptions, credits or per-clip pricing, which can come out higher or lower.

Where this comparison comes from

Specs and list prices come from OpenRouter's public video model catalogue and are refreshed daily. Makers sell the same models in their own apps and on other platforms at different prices, so treat the price column as a reference point. Every model links to its family page, where news, releases and papers about it are collected.