Same model, different bill: Nano Banana Pro, Seedance 2.5 and Veo across providers
The model is the same. The weights, the output, the quality settings. The price is not. Here is what a few models people actually use cost on each provider Flowgen can connect to.
The numbers
Each figure is the provider's list rate at default settings: 1K for images, the default resolution for video. fal's were read on 19 September 2026, Google's on 18 September, Kinovi's on 16 October. The live version of this table, refreshed daily, is at /prices.
| Model | Cheapest | Dearest |
|---|---|---|
| Nano Banana Pro, per image | Kinovi $0.067 | fal.ai $0.15 |
| Nano Banana 2, per image | Google $0.067 | fal.ai $0.08 |
| Seedance 2.5, per second | Kinovi $0.272 | fal.ai $0.473 |
| Veo 3.1 Fast from an image, per second | Google $0.10 | fal.ai $0.15 |
What that means for a real job
A set of 40 product images on Nano Banana Pro is $2.68 on Kinovi and $6.00 on fal. Ten five-second Seedance 2.5 clips are $13.59 on Kinovi and $23.65 on fal. Neither is a fortune. Both are the kind of gap you would notice on an invoice every month.
Why the gap exists
fal is a general-purpose host with a huge catalog, fast queues and good uptime, and it prices for that. Kinovi resells a smaller set of the same models at a lower rate. Google sells its own models direct. None of this is a secret, it is just tedious to compare, because every provider writes its prices in a different shape: per image, per megapixel, per second, per credit.
The catch
- A cheaper host is not always the better one. Uptime, queue time and which settings it exposes differ. Kinovi has no cancel route, so stopping a run there still bills it.
- Promotions change the order without warning. Higgsfield had 15% off several video models in September.
- Some prices are unknowable ahead of time. fal bills GPT Image 2 by the token, so no host can quote it in advance.
How I use it
Flowgen takes a key from each provider and lets every node pick its own model and host. I draft on the cheapest host, and switch the final render to whichever one has the settings I need. The model picker shows what the other hosts charge for the same model, so the switch is one click.