Five Video Configurations Ranged From 4 to 40 Credits in Our Records
Five tested video configurations cost between 4 and 40 credits, with duration, resolution, audio, and task type kept visible.

The five video configurations in my records ranged from 4 credits to 40 credits.
That range is real. It is also easy to misuse.
The cheapest entry was a veo3_1_lite preflight for a 720p, 4-second clip without audio. The most expensive was an 8-second marketing_studio_video job at 720p with a hook and setting. Those were not interchangeable jobs. The models, durations, resolution settings, audio inputs, and task types were different.
When somebody asks me what AI video generation costs, I no longer answer with a model name and a single number. I show the complete configuration.
The five configurations
| Model or task | Configuration tested | Recorded cost |
|---|---|---|
veo3_1_lite |
720p, 4 seconds, no audio | 4 credits |
cinematic_studio_video_v2 |
Standard, 4 seconds, intimate genre | 6 credits |
kling3_0 |
Standard, 5 seconds, no added parameters | 10 credits |
sync_so |
Lip sync, 716 × 1284, 7 seconds, video and audio, bounce mode | About 32.5 credits |
marketing_studio_video |
720p, 8 seconds, hook and setting, Interview plus Nature | 40 credits |
The first three figures came from cost preflights using get_cost: true. The Marketing Studio figure came from a completed smoke test. The Sync Lipsync figure came from the balance movement across two jobs. The balance went from 3464.95 to 3399.85, a total change of 65.10 credits across those two jobs.
That distinction matters. The first three rows record preflight costs. The Marketing Studio row records a completed job. The Sync row uses the balance delta across two jobs. I keep those evidence types visible instead of pretending every row was measured the same way.
The low number is not a universal model price
The 4-credit veo3_1_lite result describes 720p, 4 seconds, and no audio. Remove those settings from the row and the number becomes misleading.
The same problem applies to the 6-credit cinematic_studio_video_v2 result. That figure belongs to a standard 4-second configuration with the genre set to intimate. It does not establish what every duration or genre would cost.
The kling3_0 preflight was 10 credits for a standard 5-second request without added parameters. A separate record for Kling 3.0 Pro also shows 10 credits charged for 5 seconds, even though the estimator had said 12.5.
I do not turn either result into a permanent rate card. I keep the quote and the charge beside each other because they did not agree.
That mismatch belongs in the budget record, even when the charged result is lower. If I plan from an estimator, I need to know whether I am recording a quote or credits that actually left the balance.
The expensive rows were different tasks
The 40-credit Marketing Studio job was not just a plain clip with a longer duration. Its recorded configuration included 720p output, 8 seconds, a hook, and a setting built from Interview plus Nature.
The sync_so row was more different still. It was a lip-sync task using a video and audio input at 716 × 1284. The video ran for 7 seconds, and the sync mode was bounce. Its cost was about 32.5 credits.
I would not use those two rows to claim that audio caused the higher price, or that duration alone caused it. Too many settings changed at once. The honest conclusion is narrower: these complete tasks occupied the expensive end of this particular set of records.
Other charged measurements reinforce that caution. Sync Lipsync cost 90 credits for a 25-second video and 16.40 credits for a 4.556-second video. Marketing Studio Video cost 40 credits for its 8-second job. Those are useful observations, but they still belong to their recorded task types and settings.
Cheap output can still be unusable
The lowest nominal price is not automatically the lowest cost of a usable shot.
A 25-second Kling Lipsync attempt was priced at 7.5 credits. It failed twice. Both failed charges were refunded. The refunds prevented the attempts from becoming a permanent credit loss, but the jobs produced no usable lip sync.
Sync Lipsync cost 90 credits for the same 25-second duration in the recorded comparison. That is a severe price gap. It is also why I keep three questions in front of me: What was quoted? What was charged? Did the job produce something I could use?
A price table that answers only the first question is not enough for production.
The same applies outside generation. Sonilo Music cost 8.5 credits for 24 seconds. Topaz cost 40 credits for 25.06 seconds at 2K and 85 credits for 56.13 seconds at 2K. A Topaz 4K job cost 83 credits.
These are different operations, but they all draw from the same balance. I need them in the same production budget even though they should not be collapsed into a single “video cost.”
What I record before approving a job
I write down the requested model, the complete configuration, the evidence type, and the cost. Duration stays visible. Resolution stays visible. Audio stays visible. A hook, setting, genre, or sync mode stays visible when it was part of the request.
That is the only way the 4-to-40-credit range remains useful. The range does not tell me which model is cheapest in general. It tells me what five specific requests cost or were quoted to cost.
Put the exact duration, resolution, audio state, and task settings beside every video price in your records today. If those fields are missing, do not treat the number as a reusable rate.
Credits spent, attempts counted, results measured. I trust numbers because they don't flatter anybody, least of all me.



