What is a video token, and why does AI video cost so much?

Q: What is a video token, and why does AI video cost so much?

A video token is the unit some models, including Seedance, bill by: roughly a patch of pixels across a few frames, so the count grows with resolution, frame count and length. That is why 720p costs about twice 480p and 1080p more than twice 720p, while audio and aspect ratio leave the price unchanged.

Billing by pixels and frames

A video model generates many small space-time patches of the frame, and providers that bill by video tokens count those patches. More pixels per frame, more frames per second or more seconds all mean more tokens. The prompt, the audio track and whether the frame is portrait or landscape barely change the count, so on Seedance they do not change the price.

The steps follow pixel counts

Arc credits per 10-second clip from the published rate card, compared with the pixel ratio
Step	Pixel ratio	Seedance 2.0 Mini	Seedance 2.0
480p → 720p	2.25×	44 → 94 (2.14×)	350 → 751 (2.15×)
720p → 1080p	2.25×	Not offered	751 → 1,689 (2.25×)

The credit steps track the pixel steps closely. Length is linear: a 5-second clip costs half a 10-second clip, give or take rounding (113 and 226 credits on Seedance 2.0 Fast at 720p).

Why it adds up

One second of 720p video at 24 fps is about 22 million pixels to generate. At list prices quoted 2026-10-17, Seedance 2.0 is $0.243/s and Seedance 2.0 Mini $0.121/s, so a minute costs $14.58 or $7.29 before retakes. A series is hundreds of minutes of generation once retakes are counted, which is why model and resolution choice matter more than anything else in a budget.

Model prices on this page are Atlas Cloud list prices quoted 2026-10-17 with its free /calculate endpoint, as published in our AI video price index. Other models are priced per second by resolution tier rather than by tokens, with similar steps. Source: Atlas Cloud model catalog and /calculate list prices (atlascloud.ai/models), checked 2026-10-17

What this means in practice

Render at the lowest resolution the release needs, and upscale the finished cut if you need more.
Do not pay extra for audio on Seedance; it is the same price.
Shorter clips are not cheaper per second, but fewer wasted seconds are.
In Arc, Seedance 2.0 Mini is 9.4 credits per second at 720p against 75.1 for Seedance 2.0.

Where Katechon Arc fits

Arc converts the token-based provider price into a fixed credit cost per clip, shown before you generate, so you do not have to count tokens.

Arc does not expose raw token counts or let you set frame rate; if you need token-level control, use a model API directly.

Sources

Atlas Cloud model catalog and /calculate list prices: www.atlascloud.ai/models, checked 2026-10-17