Text to video API: what the three models cost and which to pick

Three text-to-video models are available on Genace behind a single endpoint and a single billing unit, so this comparison is exact rather than approximate.

	Seedance 2.0 Fast	Kling 2.0 Standard	Veo 3
Credits per second	5.0	7.0	12.5
Duration	any integer 3–15s	5s or 10s only	8s, fixed
Resolution tiers	720p (half price), 1080p	1080p only	720p (half price), 1080p
Aspect ratios	5	3	3
Default clip cost	25 credits (5s)	35 credits (5s)	100 credits (8s)

Picking on constraints, not on price

The per-second numbers above make Seedance look like the obvious answer, and for cost-sensitive batch work it is. But the duration column decides more pipelines than the price column does:

• Variable-length output → Seedance. It is the only model that takes an arbitrary integer duration, so you pay for the seconds you asked for. The other two round you into fixed buckets.
• Exactly 10 seconds → Kling. Seedance can do it too (50 credits vs Kling’s 70), but Kling is a single documented option rather than a number you picked.
• Anything shorter than 8 seconds on Veo 3 → not possible. durationSec is a literal 8 in the schema; there is no cheaper short-clip path.

The 720p lever

Both Seedance and Veo 3 halve their price at 720p:

	1080p	720p
Seedance, 5s	25 credits	13 credits
Veo 3, 8s	100 credits	50 credits

If you are iterating on prompts and only rendering finals at full resolution, this is a straight 50% cut on the expensive part of the loop. Kling has no 720p tier.

Image-to-video

All three accept imageInputs — an array of URLs whose first entry is used as the starting frame. Pricing is identical whether you start from text or from an image.

What one dollar buys

At the Enterprise pack rate ($0.0100 per credit):

Model	Seconds per $1
Seedance 2.0 Fast, 720p	40s
Seedance 2.0 Fast, 1080p	20s
Kling 2.0 Standard, 1080p	~14s
Veo 3, 720p	16s
Veo 3, 1080p	8s

These are totals, not single clips. Seedance caps at 15 seconds per call and Veo 3 at 8, so the longer figures mean several clips rather than one long render.

Long renders and your agent

Video generation takes minutes, not milliseconds. A synchronous HTTP call is the wrong shape for that, and it is the wrong shape twice over inside an agent loop where the session can end before the render does. Genace runs these as async jobs — see calling a video generation API without blocking your agent.

What this page does not claim

These are Genace’s prices and the schema constraints Genace enforces. Each vendor sets its own direct pricing. No claim is made here about which model produces better output — that is not something this page measures.

Models covered on this page

• Seedance 2.0
• Kling 2.0
• Veo 3