Recast or reshoot: break-even for ad variants, with Python

Recasting beats reshooting when your cost per reshot variant is more than the Recast cost of the same clip, and the Recast cost is easy to compute: source seconds times $0.375 at 768p or $0.5625 at 1080p on Sume, plus any rerenders (Video Router docs and fal, read 2026-10-03). A 15-second variant is about $5.63 at 768p. Whether that is cheaper than a reshoot depends on a number only you know, which is what a reshoot costs you.

This post gives the formula and a function. It deliberately contains no estimate of production costs, because anything it assumed would be invented.

The formula

Let S be the source length in seconds, r the per-second Recast price through Sume, a the average attempts per approved variant (including the first), and R your all-in cost of reshooting one variant. Recast cost per approved variant is S times r times a. Recast is cheaper whenever that is below R. For N variants, the totals are N times each, so the comparison does not change with N unless setup is shared.

Shared costs do change it. A reshoot can carry a fixed setup cost F (crew, location, booking) that is paid once, while Recast has none beyond the source shoot. Reshoot total is F plus N times the marginal reshoot cost; Recast total is N times S times r times a. Recast wins from the first variant when F is large, and from some N when F is small.

A function you can run

Fill in your own numbers. The defaults below are placeholders to show the shape of the output, not benchmarks.

def recast_cost(seconds: int, resolution: str = "768p", attempts: float = 1.0) -> float:
    per_second = {"768p": 0.375, "1080p": 0.5625}[resolution]
    return seconds * per_second * attempts

def break_even(seconds, reshoot_marginal, reshoot_setup=0.0,
               resolution="768p", attempts=2.0, max_n=100):
    """Smallest N where recasting N variants is cheaper than reshooting N."""
    unit = recast_cost(seconds, resolution, attempts)
    for n in range(1, max_n + 1):
        if n * unit < reshoot_setup + n * reshoot_marginal:
            return n, round(n * unit, 2), round(reshoot_setup + n * reshoot_marginal, 2)
    return None

# placeholders: replace with your own figures
print(break_even(15, reshoot_marginal=40.0, reshoot_setup=0.0))
print(break_even(15, reshoot_marginal=3.0, reshoot_setup=0.0))

How to read the result

The first call prints (1, 11.25, 40.0): with two attempts per variant, one 15-second Recast costs $11.25, which is under the placeholder $40 reshoot, so Recast wins from the first variant. The second call prints None, because a $3 reshoot beats $11.25 at any N. Neither says anything about real shoots; they show that the answer flips on your cost, so enter yours.

A worked comparison

Take a 15-second ad you want in eight versions. At two attempts per approved variant and 768p, Recast costs 8 times $11.25, or $90, and the same eight at 1080p cost $135. That is a small fraction of almost any production day, which is why the question usually is not cost but fit: do the variants differ only in the person? If the answer is yes, the budget case is easy. If the variants also need different locations, products or scripts, Recast covers only part of the work and the rest still needs a shoot or a generation step.

The formula also shows when not to bother. For a 30-second clip at 1080p and three attempts, one variant is $50.63. If a reshot variant costs you less than that, the AI route is the more expensive one. Always run the numbers with your own marginal cost before you commit a campaign to either path.

What the formula leaves out

Quality is the big omission. A reshoot gives you a new performance, new lighting and a new location, while Recast gives you the same take with a different person in it. If the variants need to differ in more than the person, Recast is not an alternative, it is a different product. There is also review time: every Recast output needs a look, and a bad tracking case sends you back for another attempt, which the attempts parameter models only crudely.

Rights are the other omission. Replacing a person with someone else's photo needs that person's agreement, and the performance in the source carries its own rights. Keep both in your approvals; Sume does not make that determination.

• Measure attempts per approved variant from your first batch.
• Use 768p for drafts, 1080p for finals.
• Compare marginal cost and setup separately.