What a redo costs in AI video, and how creative memory cuts it

In AI video, the expensive part is not the shot you keep. It is the shot you make again.

A clip with the product cropped, the wrong colour on the wrong thing, or a lighting setup your team stopped using months ago costs exactly as much as a good one. Then you pay again for the fix. Do that across a campaign and the redo loop becomes a real line in the budget.

This post puts numbers on that loop, using Blooper’s published prices, and shows where creative memory fits.

What a shot costs

Blooper prices everything in credits. One credit is one cent. The current numbers, from our pricing:

• A video clip at the default settings: 162 credits ($1.62).
• A still image at the default settings: 15 credits ($0.15).
• The Pro pack: 10,000 credits for $100. The Studio pack: 27,500 credits for $250.

Default settings means what Blooper submits when you change nothing. Longer clips, higher resolutions and other models cost more or less; each model’s price starts from its own floor, shown on the pricing page.

So at default settings, a Pro pack buys about 61 clips (10,000 ÷ 162 = 61.7). Every redo is one of those 61.

A worked example

The numbers below are illustrative. They use real prices, but the miss rate is an example, not a measurement.

Say your team makes a six-clip social cut for a client each week.

Without memory. Every new chat starts blank. Two of the six clips come back with something the client has already rejected before: the bottle cut off at the bottom, say, and a fill light the team stopped using. Each needs a second take.

• Redo cost per cut: 2 × 162 = 324 credits ($3.24).
• Over four cuts a month: 4 × 324 = 1,296 credits ($12.96).
• That is about 13% of a Pro pack, spent on mistakes that were already known.

With memory. “The product is never cropped” and “night scenes keep one light” are on your team’s memory page. The agent reads them before it writes the prompt for each shot. If that removes one of the two misses per cut, you save 162 credits a cut, or 648 credits a month. If it removes both, you save the full 1,296.

The credits are not the whole cost. Each redo is also a round of waiting, reviewing and writing the correction again. That cost is real, and it does not show up on an invoice.

Catch it at the still, not the clip

Memory works best together with a habit many of our users already have: approve the stills before anything moves.

A wrong still costs 15 credits. A wrong clip at default settings costs 162. So a mistake caught at the still stage costs about a tenth of the same mistake caught as a clip (15 ÷ 162 = 9.3%).

For the six-clip cut:

• Six stills to judge first: 6 × 15 = 90 credits.
• Redo two stills that miss: 2 × 15 = 30 credits.
• Then six clips from the approved stills: 6 × 162 = 972 credits.

Total: 1,092 credits. Skip the stills and redo two clips instead: 972 + 324 = 1,296 credits. Catching the misses at the still saves 204 credits on one cut, again with an illustrative miss rate.

Memory makes the still stage cheaper still, because the stills come back closer to what the team wants the first time. And if you always approve stills first, Blooper can learn that too, as a personal note about how you like to work.

Rules stop the call before it is priced

Notes are advice the agent weighs. Rules are stricter.

A rule is a setting your team has decided: an aspect ratio, a resolution, a model. Every generation the agent makes is checked against your rules before it is priced. If the agent picks a value that breaks one, that call is refused and the agent tries again with the right value. The wrong clip is never made, so it never costs anything.

Rules can be set for the whole team or for one client, and a rule set inside a project comes first. If a client always needs 9:16, making that a rule for the client means nobody spends a clip finding out the hard way.

What learning costs

Nothing, in credits. When Blooper reads one of your corrections and decides whether it is worth keeping, that step runs on Blooper’s account, not yours. Spotting a setting you keep repeating is plain counting, with no model involved.

The memory is also small on purpose. Only the notes that apply to the current project, client and person reach the agent, newest first, within a fixed size. Old notes do not pile up in every request.

How we will know it works

We are not going to quote a savings percentage we have not measured.

What we do instead: in about one turn in ten, Blooper leaves out one learned preference that would otherwise have applied, chosen at random, and records that it did. Comparing those turns with the rest shows whether a learned note actually changes how often a shot has to be redone. Notes you wrote, edited, or confirmed from a brand book are never left out, and neither are facts.

When there is a measured number, we will publish it. Until then, the arithmetic above is the honest version: a redo costs a known number of credits, and every redo memory prevents is that many credits you keep.

Where it is today

Creative memory is in early access. It runs on Blooper’s own team first, and a team admin switches it on for their team when it reaches them. If you want it on yours, tell us.

How memory works is in the first post of this series. The time it saves is in the third.

Start a project in Blooper, or compare credit packs.