Skip to main content

Insights

Costing AI animation by the footage you can use

Generated seconds and selected seconds tell different cost stories. A closer look at candidate clips, human handling, and early tests in an AI animation budget.

7 min read

One person hands an envelope to another. It takes only a few seconds and perhaps a single line in a storyboard. Making the shot work means getting the hands to meet, keeping the envelope intact, matching the characters’ orientation, and leaving enough time for the viewer to register the exchange.

A five-second generation does not necessarily give the editor five usable seconds. The character may look right at the beginning while the prop disappears later. The action may work while the expression fails to follow the previous shot. Someone has to try again, select a section, repair it, or find another way to stage the moment.

When estimating an AI-assisted animation, we therefore want to understand the cost of usable footage: the attempts and human work required to make a shot fit the story. The price of a generation matters. It answers one part of that question.

Five seconds can mean different denominators

Consider a small worked example. Six attempts at the same action each return five seconds and each incurs an actual charge of CNY 3. We select two seconds from the second attempt and three from the fifth, making the five-second shot we need. For this example, their continuity has been checked separately.

The total generated duration is thirty seconds, and the generation spend is CNY 18. Dividing by generated duration gives CNY 0.60 per second. Dividing by the five selected seconds gives CNY 3.60. Both calculations are valid, but the latter allocates the unused attempts to the footage that was selected.

Six five-second candidate clips, with two selected sections forming a five-second shot
Figure 1. Hypothetical candidate and selected durations. Gray marks unused intervals; orange marks the two- and three-second selections. The bottom track joins them after a separate continuity check. Allocating CNY 18 over thirty generated seconds or five selected seconds produces different unit costs. Neither includes labor.

CNY 3.60 still cannot be called the cost per finished second. Sound, subtitles, editing, repairs, and export checks are absent. Other shots may be harder. Even these five seconds could be shortened or replaced when they enter the full edit.

It helps to keep three durations separate: what the model returned, what passed a shot-level review, and what remained in the final edit. The first supports a check of tool consumption. The second helps assess a particular production task. The third corresponds to the running time delivered.

Unused footage is not always wasted work. Comparing styles, framings, or performances may help a team decide how to make the later shots. That is exploration. Repeating a failed prop interaction without improving its reference material is a different kind of effort. A single acceptance ratio mixes the two and offers little guidance on where the next part of the budget should go.

A rejection reason becomes useful when it identifies a specific problem: when the envelope disappears, where the character turns the wrong way, or why the cut from the previous shot feels abrupt. It can suggest a different input, a simpler action, or another edit. “Not good enough” rarely explains why the next attempt should improve.

A retry also involves someone’s work

While a model runs, the producer may work on other shots. Alternatively, they may have to monitor the task and submit it again. Those situations affect labor and delivery time differently. Waiting should not automatically be counted as hands-on work. Reference preparation, review, comparison, file handling, and reassembly should not disappear from the estimate simply because they happen in small pieces.

Record time against the work performed rather than the applications opened. A clip may move from an image tool to a video tool and then into an editor, with comparison and repairs between them. One person’s twenty minutes should be counted once. Two people working for twenty minutes contribute forty person-minutes. Neither figure directly states the elapsed project time.

Now expand the example to a thirty-second production exercise: twelve planned shots, four attempts per shot on average, and CNY 3 per attempt. Assume twenty person-hours at an internal costing rate of CNY 100 per hour, plus CNY 300 for assets and software allocation. That last amount does not include the generation charges already listed.

ItemAssumption in this exampleAmount
Labor20 person-hours × CNY 100CNY 2,000
Generation12 shots × 4 average attempts × CNY 3CNY 144
Other direct costsFixed asset and software allocationCNY 300
TotalSum of the three items aboveCNY 2,444

Preparing a customer quote also requires considering tax, equipment, management, sales, margin, and project risk under the team’s own costing scope.

Varying average attempts and labor hours while holding the other inputs fixed shows how the budget changes under different combinations.

Hypothetical budget heatmap for twelve, twenty, or twenty-eight labor hours and two, four, or eight average attempts
Figure 2. Sensitivity of the hypothetical budget. Twelve planned shots, CNY 3 per attempt, CNY 100 per person-hour, and CNY 300 in other costs remain fixed. Each cell shows a total; the outlined cell is the CNY 2,444 baseline. The axes vary independently and do not predict labor hours from attempt counts.

At the baseline, cutting the price per attempt from CNY 3 to CNY 1.50 saves CNY 72 over forty-eight attempts. Four additional person-hours of selection and repair add CNY 400. If both changes occur, the total becomes CNY 2,772—CNY 328 above the original budget.

Comparing tools requires looking at the work needed to finish the same shot. One may cost less per call but need more preparation. Another may charge more but suit a particular action better. Comparable task records are needed to tell which is economical; a compelling demonstration or a price list is not enough.

After a few projects, separate the shot types too: a stationary character, a two-person conversation, hand-object contact, and a moving group can demand different preparation and handling. An average derived from easy shots is a weak basis for estimating the whole production. A few demanding requirements may account for much of the uncertainty.

Put a demanding shot near the beginning

When a budget is tight, exposing an important difficulty early can be a useful investment.

Arrange rough drawings and temporary sound on a timeline. Check whether the story reads, whether the action has enough time, and whether dialogue crowds out a reaction. Adobe’s introduction to animatics explains this kind of timed storyboard preview. The drawings can remain rough while the intended meaning becomes clear.

Then test a shot that represents the actual difficulty. If the challenge is two characters exchanging a prop, test the exchange. If it is a character moving from a close view into a wide shot, test appearance and spatial continuity. A successful stationary portrait does not establish the budget for those actions.

Reference controls can help prepare such tests. Google’s October 15, 2025 Veo 3.1 announcement described reference-image guidance and generation between first and last frames. These offer ways to specify characters, scenes, and action endpoints. The number of attempts and amount of post-production a particular shot needs still have to come from the team’s own tests.

The test makes the next decision more concrete. An action may deserve further work because the story depends on it. A transition may communicate the same information through a different framing, a reaction, and sound. Some requirements may need to be reduced to fit the budget. The aim is to remove unproductive repetition while preserving what the audience must understand.

An exploration allowance for a new character, style, or action is reasonable as well. Reference material and handling experience from the first effort may help later shots. Reuse depends on how closely the clothing, setting, and task match; future reuse should not be booked as a saving before it happens. A review point for each kind of exploration gives a stronger basis for decisions than a universal promise about a maximum number of attempts.

The quote should also state what will be delivered: duration, aspect ratio, sound and subtitle versions, whether editable project files are included, and which changes to agreed material require a fresh assessment. Clear conditions help people understand the work behind “make a video.”

We want AI to help small teams put more ideas on screen. Sustaining that work means seeing the effort beyond the selected images: unused attempts, human judgment, sound, editing, and the consequences of a revision. Useful cost knowledge makes the next comparable shot easier to plan: what to prepare, where more effort is justified, and when to try a different approach.

View all