Most AI video budgets start from a plan's headline credits and stop there. The number a finance lead needs is different, and it takes a few more lines to get right.
The method applies to subscription and credit-metered tools such as Runway, HeyGen and Synthesia, for any deliverable you can judge against a written brief. The result is only as good as your attempt log. Vendor per-minute estimates vary with model, avatar type and length, so measure your own rate over a few weeks before you set a budget.
What Goes Into The Cost Of An AI Video
The cost of an AI video is everything spent to reach one accepted deliverable, meaning the credits on every attempt, the hours spent prompting, reviewing and editing, and the credits that expire unused, and with more than a third of video professionals using AI according to Wistia's State of Video survey of 900+ professionals, that total belongs in the budget.
Wistia's 2026 report also found that nearly a quarter more plan to start soon, and over half are putting more money into AI this year. Once AI video generation tools are a normal part of the workflow, they need a cost line that holds up when someone asks what each video cost.
The unit matters because every other number divides by it. If the brief is vague, "accepted" drifts, and so does the cost.
The cost of one accepted video has five lines:
- Credits on every attempt. Every completed generation, kept or rejected, uses its full credits.
- Top-up or add-on credits. Credits bought on top of the plan when the monthly allowance runs out.
- Credits that expire unused. On several plans, credits left at the end of the billing period are lost, which raises the real cost of every credit you did use.
- Human hours. Brief and prompt writing, review, editing and captions, and approval rounds.
- Allocated assets and seats. Any stock, music, voice or seat cost that exists only because of the video work.
Credits alone mislead for two reasons. Runway, for example, charges for a completed generation even when it misses the brief, so the credits on the clip you keep are only part of the spend. And the people time around each video is usually priced at zero in a credit calculator, even though someone is paid for every hour of it.
How To Calculate The Cost Of An AI Video
To calculate the cost of an AI video, add the money value of the credits spent on every attempt to the priced hours of prompting, review and editing, add any credits that expired unused, then divide by the number of videos that passed the brief.
In words, the formula is credits used times cost per credit, plus hours times a loaded hourly rate, plus allocated expiring credits and assets, all divided by accepted videos. As a single line you can paste into a worksheet note:
cost per accepted video = ((credits on all attempts x cost per credit) + (hours x loaded hourly rate) + expired credits and assets) / accepted videosThe steps below fill each part of that line. Work through them in order for each video or campaign.
- Write the brief and the acceptance test before the first generation, so "accepted" means the same thing to everyone who reviews.
- Open an attempt log with one row per generation, including the ones you reject.
- Record the credits each attempt used, taken from the tool's own usage log rather than estimated.
- Convert credits to money at your effective cost per credit, which is the plan fee divided by credits actually used, or the top-up price for bought credits.
- Log human hours per stage, covering brief and prompt writing, review, editing and captions, and approval.
- Price those hours at a loaded hourly rate, using the same rate for every option you compare.
- Add the share of credits that expired unused in the period, plus any assets or seats bought for the work.
- Divide the total by accepted videos only, never by every clip generated.
Why the meter needs your own log: Runway charges credits per second at a rate that differs by model, so a single generation's credits depend on its length. Gen-4.5 uses 12 credits per second, which makes a 5-second clip 60 credits and a 10-second clip 120 credits. Two attempts on the same model can cost different amounts if their lengths differ, which is why each row records seconds as well as credits.
How to total credits per video: In Google Sheets, SUMIF adds the values in one range where another range meets a condition. With video IDs in column A and credits in column E, this returns the credits spent on one video across every attempt:
=SUMIF(A2:A500,"{{video-id}}",E2:E500)Replace {{video-id}} with the ID in the log, or point the criterion at a cell that holds it. The same pattern totals hours per video when the hours sit in their own column.
When nothing passes: If no attempt passes the brief, there is no cost per accepted video yet, only spend. Carry the spend forward to the video that eventually passes, or record it as the cost of a failed brief. Either way it stays in the numerator.
The worked example further down follows these same steps for two deliverables, so you can see where each number lands.
How Do Failed Generations Affect AI Video Cost
Runway returns credits automatically only when a generation ends in an error, so a clip that finishes but misses the prompt or loses a likeness is charged in full and belongs in the cost of the video you eventually keep.
That splits "failed" into two very different cases.
- A generation error. The job fails. On Runway, credits from an errored generation reappear within a few minutes. If the balance does not rise, the generation most likely failed before any credits were charged.
- A completed generation that misses the brief. The clip renders but does not match the prompt, does not preserve a likeness or is otherwise unsatisfactory. Runway consumes the credits and does not automatically reinstate them.
The costing rule follows directly. Every completed attempt goes in the numerator, and only accepted ones go in the denominator. Errors that refunded their credits can be logged with zero credits, so the attempt count stays honest without double-charging.
Editing is the other way to fix a miss, and on Synthesia it can cost less than a fresh generation. Synthesia charges an edit to an existing video only for the duration it adds. New words, replaced words, an added or replaced avatar, or a changed voice language or accent are charged. Removing words, reordering words, punctuation, avatar framing, removing an avatar, changing the avatar look, pronunciation and voice tone changes have no credit impact. Credits are recalculated on the updated version.
Decision rule: Regenerate when the shot is wrong, and edit when the words are. A wrong camera move or a lost likeness needs a new generation. A mispronounced product name, a sentence in the wrong order or a stray comma is an edit, and on Synthesia that edit is free.
Do not plan around a typical failure rate. Your own attempt log is the only rate that applies to your briefs, your models and your reviewers.
How To Turn Subscription Credits Into A Cost Per Second
Runway's Gen-4.5 model uses 12 credits per second of generated video, so every 10-second attempt costs 120 credits whether you keep it or not, and on the Standard, Pro and Unlimited plans any monthly credits left on the billing date expire. Your cost per credit is what you paid divided by credits used.
Each tool meters differently, so the conversion starts with knowing the unit.
How each tool meters credits
| Tool | Meter | Example | Unused credits |
|---|---|---|---|
| Runway | Per second, by model | 120 credits per 10s Gen-4.5 clip | Expire monthly on Standard to Unlimited |
| HeyGen | Per minute or per generation | 31 credits per Avatar IV Video Look minute | Roll over one cycle on monthly plans |
| Synthesia | Shared credits across features | 1,000 to 2,000 per minute, typical | Reset each billing period |
Runway
- Meter. Runway bills per second at a rate set by the model, with Gen-4.5 at 12 credits per second.
- Failed generations. Credits come back only after a generation error, never for a completed clip you reject.
- Unused credits. On Standard, Pro and Unlimited, monthly credits expire on the billing date. On Max, up to one month of unused credits may roll over, and on Team up to one month of unused pooled credits may roll over.
- Extra credits. Purchased credits do not expire, so they can carry a quiet month without loss.
For plan-by-plan detail, see Runway's plans, credits and top-up rules.
HeyGen
- Meter. HeyGen meters some features per minute of output and others per generation. Avatar III Video Look uses 4 credits a minute, Avatar IV Photo Look 16, Avatar IV Video Look 31, Avatar V 48 and Custom Expressive Motion 40. Video Translation uses 4 to 10 depending on mode, and Video Agent 80 in standard mode or 240 in Seedance mode. AI model training is 60 credits per training and image assets 2 credits per image. HeyGen says usage may vary by feature, model, duration or generation type.
- Unused credits. Monthly subscribers' unused credits roll over for one additional billing cycle. Annual subscribers' unused credits accumulate until the annual renewal date. Cancelling a paid plan loses unused credits, and HeyGen deducts the credits closest to expiring first.
- Extra credits. On Business, extra credits cost $0.05 per credit, bought in blocks of 100 for $5, with auto-reload on Business only. Creator and Pro do not support one-time credit purchases.
Synthesia
- Meter. Synthesia credits are one shared currency across its usage-based features, so video, dubbing and surveys draw from the same balance. Its pricing page says a one-minute video typically uses 1,000 to 2,000 credits and that actual usage varies.
- Unused credits. Self-serve credits do not accumulate. They reset at the start of each billing period, and unused credits from the previous plan do not carry over when you upgrade.
- Edits. Edits are charged only for the duration they add, and many edits are free, as the failed-generations section above sets out.
Synthesia's plans are shown here:
Synthesia pricing
| Plan | Price | Billing | Notes |
|---|---|---|---|
| Basic | $0 | per month | |
| Starter |
|
| |
| Pro |
|
| |
| Enterprise | Custom pricing |
How to get your effective cost per credit: Divide what you paid in the period by the credits you actually used. If you paid for a full allowance and used half, each used credit cost twice its headline value, which is how expired credits raise the cost of every video. Bought top-up credits are priced at their top-up price.
Where to take the meter from: Vendor figures such as Synthesia's 1,000 to 2,000 credits a minute are averages the vendor itself says vary. After a few weeks of logging, divide your credits by your seconds or minutes per tool and feature. That measured rate is the one to budget with.
How To Price Editing Time For AI Video
Price editing and review time at a real hourly rate, and when a team has none, the US Bureau of Labor Statistics puts median pay for film and video editors and camera operators at $36.10 an hour in May 2025, before overhead. Use your own loaded rate when you have one.
That BLS figure is a median wage for the combined group, or $75,100 a year. A loaded rate adds overhead such as benefits, tools and management time, so it sits above the wage. The wage is a floor for a benchmark, not a ceiling.
Time each stage separately, because each one moves for a different reason:
- Brief and prompt writing. Time spent turning the brief into prompts or a script.
- Review. Watching each completed attempt against the acceptance test.
- Editing and captions. Trims, captions, overlays and any edits made inside the tool.
- Approval rounds. Stakeholder review and the changes it triggers.
Leave out waiting time while a render runs, unless the person cannot do other work in the meantime. Counting it inflates the hours line without telling you anything about the brief.
The same rule answers whether AI video is cheaper than hiring an editor. Price the editor's hours and the AI workflow's hours at the same rate, add the AI workflow's credits, and compare cost per accepted video on the same brief. Any other comparison mixes units.
Edit rules shift work between the two lines. On Synthesia, removing or reordering words, fixing pronunciation or changing voice tone costs no credits but does cost the minutes someone spends on it, while added words cost both.
Decision rule: If editing hours times the rate exceed the credit spend on a video, fix the brief before switching tools. A tighter brief cuts review and editing on any tool, while a new tool keeps the same loose brief.
How To Track Cost Across An AI Video Workflow
Track AI video cost in two places, hours in Toggl Track, where labour costs need the Premium or Enterprise plan and only Administrators and Project Leads can set or see them, and credits in a Google Sheets attempt log with one row per generation.
Hours in Toggl Track: In Toggl Track, create one project per video or per campaign, then use tasks or tags for prompt, review, edit and approval. Two features turn those hours into money.
- Billable rates. A Starter plan feature applied to time-entry duration. Rates can be set at workspace, workspace member, project, project member and task level, and the most granular rate applies. On Starter, changing a rate changes it for all past data. On Premium and Enterprise, past reports are not altered. Billable rates in reports are visible only to admins unless the "Who can see Billable Rates" setting is changed to Everyone.
- Labour costs. An hourly cost per workspace member or per project member, with the project level overriding. Labour costs cannot be assigned to user groups, are available on Premium or Enterprise only, and can only be defined and viewed by Administrators and Project Leads. The Project Lead role is Premium only. The Profitability Report uses project-level labour costs when they exist.
For costing, labour costs are the closer fit, because they hold what an hour costs the team rather than what it is billed at.
Credits in Google Sheets: In Google Sheets, build the attempt log with these columns:
- video ID
- tool
- model or feature
- seconds
- credits
- status
- hours
Make the status column a dropdown so every row reads the same. Select the cells, open Dropdowns, then set the options on the Data validation rules panel to accepted, rejected and error. SUMIF then totals credits per video ID, and a second SUMIF on the status column counts what was accepted.
What to record, where, and what it needs
| What to record | Where | Plan, permission or note |
|---|---|---|
| Billable rate per hour | Toggl Track billable rates | Starter plan, admin-only view by default |
| Hourly labour cost | Toggl Track labour costs | Premium or Enterprise, Admins and Project Leads |
| Profit by project | Toggl Track Profitability Report | Uses project-level labour costs |
| Credits per attempt | Google Sheets attempt log | One row per generation |
| Attempt status | Sheets dropdown via data validation | Accepted, rejected, error |
| Automated HeyGen runs | HeyGen API credits | Separate API key and credits |
Who sees what: Decide before launch who needs rates and costs. If reviewers log hours but should not see pay data, keep labour costs with Administrators and Project Leads and share only the per-video totals from the sheet.
A Worked Example With An Avatar Video And A Generated Clip
A 10-second Runway Gen-4.5 shot that takes four attempts to get one keeper consumes 480 credits, four times the 120 a single attempt suggests, before any review or editing hours are added.
The two examples below use the meter rates above. Attempts, accepted count, hours and cost per credit are example inputs, so replace them with the numbers from your own log. The two are priced side by side because avatar video and generative video bill differently, one per minute of presenter output and the other per second of generated footage. To compare meters across AI avatar video tools, run Example A's inputs through each one.
Example A, a one-minute HeyGen avatar explainer. Avatar IV Video Look at 31 credits a minute, with one rejected render and one accepted. Hours are 0.5 for script, 0.5 for review, 1 for editing and captions and 0.5 for approval.
Example B, a 10-second Runway product shot. Gen-4.5 at 120 credits per 10-second attempt, with four attempts and one accepted. Hours are 0.5 for prompts, 0.5 for review and 0.5 for editing.
Both use $0.05 per credit, the HeyGen Business top-up price, as the example cost per credit, and the BLS median of $36.10 an hour as the example rate.
Example inputs, not measured rates
| Line | A, HeyGen avatar | B, Runway clip |
|---|---|---|
| Attempts | 2 | 4 |
| Credits per attempt | 31 | 120 |
| Total credits | 62 | 480 |
| Hours | 2.5 | 1.5 |
| Hours cost at $36.10 | $90.25 | $54.15 |
| Accepted videos | 1 | 1 |
| Cost per accepted video | $93.35 | $78.15 |
Three points come out of the arithmetic.
- Rejected attempts multiply the credit line. Example B's credit cost is four times a single attempt, because three completed clips missed the brief.
- Hours drive both totals at a professional wage. In Example A, hours are $90.25 of $93.35. In Example B, they are $54.15 of $78.15.
- Expired credits would add to both. Neither example includes credits lost at the end of the period. If half the allowance expired, the effective cost per credit would double, and so would the credit lines.
Your effective cost per credit will differ from $0.05, especially on a plan allowance rather than top-ups. Use the figure from your own plan fee and credits used.
How To Lower The Cost Per AI Video
A direct way to lower the cost of an AI video is to raise the share of attempts that pass the brief, because on a per-second meter such as Runway's every rejected 10-second attempt costs the same credits as the one you keep.
Each lever below works on one line of the formula.
- Raise the accept rate. A tighter written brief means fewer completed clips that miss it. Test prompts at the shortest duration before full length, because per-second meters charge by length. On Runway Gen-4.5, a 5-second test is 60 credits against 120 for a 10-second attempt.
- Edit instead of regenerating. Where an edit is free or charged only for the duration it adds, as on Synthesia, fixing the words costs less than a new render.
- Spend expiring credits first. Plan heavy generation for the days before a billing date on plans where credits expire. HeyGen already deducts the credits closest to expiring first. Matching the plan to measured volume stops the expired-credit line from growing.
- Put AI where teams already use it most. Wistia's 2026 report found AI is most common in pre-production, meaning planning, scripting and ideation, and that captions top the list of generative uses for video assets, with dubbing next. Settling the script before the first generation keeps paid attempts for the shot, not for a script that is still changing.
- Set the acceptance test from repeat runs. Scoring several outputs of the same brief shows how much variation to expect before a reviewer rejects a clip. A repeatable study design for brand consistency sets out how to run and score those repeats.
Common Mistakes When Costing AI Video
A costing mistake that skews every comparison is dividing credit spend by every clip generated rather than by the clips that passed the brief, which makes a tool look cheaper the more attempts it needs.
The others follow the same pattern of leaving a real cost out.
- Using headline credits. Dividing the plan fee by the credits a plan includes, rather than credits used. The fix is to divide by credits used in the period.
- Losing credits on a Synthesia upgrade. Unused credits from the previous plan do not carry over when you upgrade. The fix is to upgrade near the end of a billing period, after the old credits are spent.
- Losing credits on a HeyGen cancellation. Cancelling a paid plan loses unused credits. The fix is to spend down the balance before cancelling.
- Treating rejected clips as refundable. Runway reinstates credits only after a generation error, not for a completed clip that misses the brief. The fix is to log every completed attempt as spent.
- Forgetting the API meter. Automated HeyGen workflows through Zapier, Make or n8n use API credits, not plan credits. The fix is to log API credits as their own line.
- Pricing hours at zero. The fix is to apply one loaded hourly rate to every stage and every option.
FAQ
Do failed AI video generations use credits?
Failed AI video generations use credits only when they complete, since Runway returns credits automatically after a generation error but keeps the charge for any finished clip, even one that misses the prompt. Credits from an errored job reappear within a few minutes, so log refunded errors at zero and completed rejects at full cost.
Do unused AI video credits roll over?
It depends on the tool and plan. Runway's monthly credits expire on Standard, Pro and Unlimited, while Max and Team can roll over up to one month. HeyGen monthly plans roll unused credits over for one extra cycle. Synthesia self-serve credits reset each billing period and do not accumulate.
Should the subscription fee count if credits go unused?
Yes. The fee is spent whether or not the credits are used, so divide what you paid by the credits you actually used to get the true cost per credit. Expired credits on Runway, Synthesia or a lapsed HeyGen cycle raise the cost of every video made in that period.
Does editing an existing AI video cost credits again?
On Synthesia, only edits that add duration cost credits, such as new or replaced words, an added avatar or a changed voice language. Removing or reordering words, punctuation, framing, pronunciation and voice tone changes are free, and credits are recalculated on the updated version, so many fixes cost time rather than credits.
Sources
Sources


