Nobody publishes their discard rate. Scroll the launch posts, the tutorials, the one-take-wonder demos, and AI video looks like a medium where creators type a prompt and download a masterpiece. Then you budget your first real project on that fantasy, drain half a month’s credits by shot four, and sit there wondering what is wrong with you. Nothing is wrong with you. You are just the first person on the project who counted honestly.
Every finished AI video has an invisible half: the generations that did not ship. This article puts a number on them. We will break down realistic retake rates by shot type, show how storyboard approval cuts the tax before it is charged, and build a credit budget worksheet for a thirty-second video — with clearly labelled illustrative numbers you can swap for your own tool’s costs. Bring your last project’s generation history if you have one. By the end, retakes stop being a shameful leak and become what they actually are: a budget line professionals plan for.
The Invisible Half Of Every AI Budget
Call it the attempts ratio: total generations consumed divided by approved shots in the export. It is the single most predictive number in a creator’s economy. More than plan price. More than credit allowance. Because it multiplies every other figure you budget with. A plan that looks ample at one generation per shot is a famine at four.
The ratio hides for social reasons. Creators post keepers. Platforms demo keepers. Even honest tutorials compress three attempts into one satisfying click, because showing the corpses is less watchable. So beginners calibrate to fiction, assume their discard pile means incompetence, and do the two worst things available: they under-budget the next project the same way, and they burn credits trying not to be the kind of creator who needs retakes. Everyone needs retakes. The pros just pre-pay for them in planning instead of mid-panic at a top-up screen.
Start logging the ratio per finished video this week. In my Seedance 2.5 and Veo work — generating from storyboard-approved frames with locked references — it is the first number I check when a project closes. Before views, before comments. A ratio creeping upward across a series is an early-warning system: a character sheet going stale, a shot type creeping past what the model does well, boards getting loose. The budget impact and the craft diagnosis arrive in the same number. Our AI video QC checklist helps here, because a shared definition of what counts as approved keeps the ratio honest instead of vibes-based.
Retake Rates By Shot Type: Faces, Products, Camera Moves
Not all shots tax you equally. Budgeting as if they do is how projects run dry precisely at the hardest shot — which always seems to schedule itself last. Think in bands rather than false precision. Your bands will differ by tool and craft; these are planning shapes built from how production actually behaves. Verify them against your own log and adjust.
Faces tax highest. A speaking or reacting close-up must hold identity, skin, eyes, and mouth across every frame, and viewers forgive nothing because evolution built them face-inspection machines. Budget the most attempts here — and reduce the risk upstream: one speaker, one small beat per shot, dialogue the model can physically say. Products come next. Edges, labels, and hands touching the object are melt magnets, so product hero shots earn a generous attempt allowance, partially offset by the fact that a stable reference frame does heavy lifting. Camera moves are the sneaky middle: a simple slow push-in is cheap when the room cooperates, while long travelling moves through detailed spaces invite the environment to redesign itself. Attempt costs climb with move ambition and environment detail. Establishing and atmospheric shots run cheapest. No faces to break, no labels to respell, motion that can be gentle. Ambient generations are where tight budgets breathe.
| Shot type | Relative retake pressure | Why it fails | Board-level prevention |
|---|---|---|---|
| Face close-up, speaking or reacting | Highest | Identity, eye, and mouth drift across frames | One speaker, one beat, locked character reference |
| Product hero with hands | High | Melting contact points, label and edge instability | Clean prop reference, slow single reveal move |
| Ambitious camera move | Medium to high | Space re-arranges, speed and direction disobeyed | Name one simple move on the board and freeze the rest |
| Medium action shot | Medium | Gesture muddle, wardrobe or prop slips | One clear action per shot, itemized outfit line |
| Establishing or atmosphere | Lowest | Generic drift rather than hard failure | Locked location frame, gentle motion only |
Weight your worksheet by this table and the budget starts predicting reality. Six shots is not six equal costs. It is two expensive shots wearing four cheaper ones as camouflage. Plan the expensive ones early in the credit cycle, while the pool is fat and your temper is fresh.
Cutting Retakes At The Storyboard, Not The Render
Here is the craft truth underneath the arithmetic: most bad generations were unapproved frames that should never have been animated. The composition was muddy. The action was two actions. The frame asked the model to invent a face, a product, and a location simultaneously. Generation did not fail the shot. The shot arrived pre-failed — and credits were spent discovering it.
Storyboard approval inverts that economics. Judge the frame as a still, where judging is free. Composition, character accuracy against the sheet, wardrobe, product integrity, light, shot size — all decidable before a single credit moves. Write the motion line on the board: one movement for the subject, one for the camera at most, and name what holds still. In our pipeline those boards drive Seedance generations directly, and the drop in attempts ratio was the single largest cost saving I have ever made in this medium. Larger than any plan change. The full method lives in our storyboard-first workflow guide. For budgeting purposes, remember the exchange rate: an hour of boarding routinely buys back a month of re-rolls.
Build a second gate into the board review — the rework trigger. Decide in advance what evidence sends a shot back to the frame instead of back to the generator. Mine is plain: after the agreed attempt budget, if the failure lives in the frame itself — wrong composition, impossible action, reference that never matched — the next spend is an eraser, not a credit. Creators resist this because regenerating feels like progress and reboarding feels like retreat. The budget knows better. The frame is where the failure is cheapest, by definition, because it is free to fix.
A Credit Budget Worksheet For A 30-Second Video
Time to fill in the worksheet. All numbers below are illustrative placeholders, labelled so you can replace them with your tool’s current credit costs at your settings — which you should verify on the generation screen before a project, not from memory.
Step one: list the shots from the approved board, with type. Say our thirty-second short has six: one face close-up, one product hero, two medium action shots, one camera move through the location, one establishing shot. Step two: assign each an attempts allowance from your bands. Illustratively: face three attempts, product three, mediums two each, camera move two, establishing one. That is a planned thirteen generations for six keepers. Step three: price the generation. Say one generation at your duration and settings costs a round ten credits in our illustration; the shot budget lands at one hundred thirty credits, plus a contingency line of twenty percent for the shot that surprises everyone. Call it roughly one hundred sixty credits, all-in, known before generation begins. Step four: sanity-check against the pool. If that figure eats an unreasonable share of your cycle’s allowance, the fix happens now, on paper: simplify the camera move, convert the face shot to a reaction over an existing frame, or schedule the project across a credit boundary deliberately instead of discovering the wall at shot five.
| Worksheet line | Your number |
|---|---|
| Shots in the approved board, by type | List each shot and its type (face, product, action, move, establishing) |
| Attempts allowance per shot | From your logged bands per shot type, not from hope |
| Total planned generations | Sum of allowances |
| Credit cost per generation at your settings | Verified on the generation screen today |
| Shot budget | Planned generations times credit cost |
| Contingency (15–25%) | For the shot that surprises you, because one will |
| Total project reserve | Shot budget plus contingency, checked against the credit pool and its expiry |
Run that worksheet before every project above a single clip, and two lovely things happen. Quotes and posting schedules stop being fiction. And the end-of-project review writes itself: planned attempts versus actual, per shot, logged against the board. That log is the asset that makes next quarter’s worksheet smarter. It is also your upgrade evidence — any tool change, plan resize, or billing decision gets argued from those rows, not from how a renewal banner made you feel. Our broader pricing breakdown shows where this worksheet plugs into monthly planning.
When Retakes Mean Rework The Frame, Not Re-Roll
The hardest budgeting skill is stopping. A shot has eaten its allowance and the last three generations failed in recognizably the same way. Same hand through the same handle. Same face sliding on the same turn. That sameness is diagnostic gold: the model is not unlucky, it is telling you the request conflicts with physics, the reference, or its training. Another roll buys another copy of the same failure at full price.
Rework options, in the order I try them: simplify the motion line to the one action that matters; strengthen or replace the reference frame so the model invents less; change the shot size, because a face that breaks in close-up often holds beautifully in a medium; or split the shot, letting the edit join two simpler generations where one heroic one kept dying. Each option spends thought instead of credits. Save re-rolls for failures that vary randomly — a good frame with unstable luck, where the next attempt genuinely rolls new dice. Learn to tell those two failure families apart and your attempts ratio drops again, because half of most discard piles are deterministic failures being gambled on.
And close the loop kindly. Some shots, after honest rework, get cut from the board — and the sequence gets better for it. A budget that cannot absorb a cut shot is not a budget. It is a hostage negotiation with your own storyboard.
Frequently Asked Questions
What Is A Normal Attempts-Per-Keeper Ratio For A Beginner?
Expect it to be humbling at first — often several generations per approved shot while you learn a model’s habits and before your frames are storyboard-approved. What matters is the trend: log it per project and watch frame-first planning pull it down over a month of work.
Should I Budget Retakes Separately For Each Shot?
Yes. Faces and product shots fail at materially different rates from establishing shots, so a flat per-shot allowance overfunds easy shots and starves hard ones. Budget by shot type from your own log, with a small project contingency on top.
When Is Another Regeneration Just Wasting Credits?
When the failure repeats in the same way across attempts. Repeatable failure means the frame, motion line, or reference is asking for something the model cannot give as specified. Rework the board rather than buying another copy of the same result.
The Takeaway
Retakes are not a character flaw. They are a production constant with a rate you can measure per shot type, shrink at the storyboard, and reserve credits for like a professional. Budget the generations you will actually spend, include the contingency, and stop re-rolling deterministic failures. Your credit pool lasts longer, your schedules become believable, and your discard pile turns into data.
Start with your next project: board it fully, fill the worksheet before generating, and log actuals against it in the edit. If you want the planning system behind those boards, our storyboard-first workflow guide is the full method.