The first number most teams ask for is the generation fee. The number that actually decides whether the pilot was worth running is the fully loaded cost of moving one short piece of work from brief to a tested revision path.
I keep a simple baseline for a small-agency Seedance 2.5 pilot so the conversation stays grounded. The numbers below are drawn from controlled tests with teams of roughly three to seven people who already understand commercial production. They are not universal. They are a starting reference that any team can replace with its own measured data after a single disciplined run.
The point of the baseline is not to predict every future job. The point is to stop treating the model’s per-second or per-generation price as the full cost of the work.
What the Baseline Includes

I measure five categories on every pilot:
Preparation labor (brief interpretation, locked list, reference package, variation budget, hypothesized note)
Generation and variation management (including failed or discarded takes)
Internal selection and marking
First realistic revision test (the pre-written note applied under controlled conditions)
Handoff documentation (the short note that would travel to an editor or finishing artist)
I deliberately exclude final finishing, color, sound mix, and actual client delivery. Those costs exist, but they belong to a later stage. The baseline answers a narrower question: what does it cost to reach a take that has survived one realistic note and is ready for the next role?
A Representative Small-Agency Pilot
The following numbers come from a composite of recent controlled tests using a synthetic thirty-second product brief, clean white-box references, a fixed variation budget of four to six takes, and one pre-agreed client-style note.
Preparation
One producer or technical lead, 45–90 minutes
Reference gathering and labeling, 30–60 minutes
Total preparation labor: roughly 1.5–2.5 hours
Generation cycle
Wall-clock time varies with service load and settings; the labor that matters is the person monitoring, reviewing thumbnails, and managing the variation set.
Active labor: 30–75 minutes for a small fixed budget
Discarded or failed takes are included in this window; they are not free.
Internal selection and marking
Two to three people, 20–40 minutes
Criteria are written in advance, so the meeting stays short.
Revision test
Application of the pre-written note (local adjustment or targeted regeneration)
Labor: 15–60 minutes depending on whether the change stayed local
This is the stage that most “generation-only” cost models omit.
Handoff note
10–20 minutes
Short, factual, and required before the take is allowed to move.
Fully loaded labor for the pilot path
Typical range I have recorded: 4–8 hours of combined team time to reach a marked take that has survived one realistic note.
Generation service cost sits on top of that labor. It is real and should be tracked, but in every pilot I have measured it has been the smaller of the two numbers once human time is counted at normal agency rates.
What Moves the Numbers
Several factors consistently push a pilot toward the high or low end of the range.
Toward the low end:
Clean, pre-existing product references
Written selection criteria and variation budget before generation
A revision note that stays local
One person who already understands the marking system
Toward the high end:
Mixed or outdated reference folders
Open-ended generation without a fixed budget
A note that forces full regeneration and a second selection cycle
Multiple stakeholders applying different private standards during selection
The model’s speed does not automatically move the pilot to the low end. Preparation and revision discipline do.
How I Recommend Teams Use the Baseline
Do not treat these figures as targets to hit. Treat them as the first measuring stick.
Run one controlled pilot with the five categories above. Record actual labor by role and actual generation cost. Compare the result to the baseline. The gaps will tell you where the process is still leaking time.
Common early findings:
Preparation was rushed and the revision test paid for it
Variation budget was too open and selection became the bottleneck
The “small” note was not small after all
Handoff documentation was skipped and the next role had to ask basic questions
Each of those findings is cheaper to discover on a synthetic brief than on a live client job.
A Short Calibration From Recent Tests
In one recent run with a five-person regional team the fully loaded path landed just under five hours. Preparation was thorough, the white-box package was clean, the note stayed local, and the handoff note took twelve minutes. Generation service cost was modest relative to the labor.
In an earlier run with a different team the same brief consumed closer to nine hours. The difference was not the model. It was an incomplete locked list, an open variation set, and a note that forced a full restart. The pictures from the second run were not worse. The process was simply less prepared.
The dog slept through both. Morgan, looking at the two time sheets, said the same thing she says about project estimates: the version that looks cheaper on day one is often the one that is still accruing cost on day three. She remains correct.
What to Measure on Your Next Pilot

Preparation hours by role
Active generation and variation-management hours
Selection and marking hours
Hours spent on the first realistic revision test
Hours spent on the handoff note
Generation service cost for the entire variation set (including discards)
Add the numbers. That sum is the real baseline for your team under your current process.
Once you have it, you can decide whether the next experiment should focus on tighter preparation, clearer revision scope, fewer but better-referenced takes, or a different stage entirely.
The generation fee is easy to quote. The fully loaded cost of a take that can survive the first note is the number that tells you whether the pilot is becoming a workflow.
If it cannot survive the handoff, it is not a workflow yet.

No letters yet — be the first to write.