AI Video Credits Explained: What 1 Credit Buys, and How to Compare Vendor Prices

A credit is whatever the vendor decides it is

Two browser tabs, two prices, and no honest way to put them side by side. One tool quotes 30 credits for a short clip. The other quotes 2,000 credits a month. Nothing in those two numbers touches. They are denominated in different currencies that happen to share a name, and neither vendor is obliged to tell you the exchange rate in a form you can divide.

Adobe documents its own version of this more plainly than most, on a page that carries a date. In the revision stamped August 5, 2026, its generative credits FAQ says that most standard generative features, like Generative Fill in Photoshop, "use 1 credit per generation." A few lines down, the video rate card prices Generate Video at 100 credits per second at 1080p and 24 frames per second. Same company, same page, same word: one credit buys a whole still image, or one hundredth of a second of video.

That is the whole problem in one document. A credit is not a unit of time, of output, or of quality. It is a token the vendor issues and redeems at a rate it writes and can rewrite. Everything a buyer wants to know sits in the conversion table, not in the balance, and the conversion table is a moving document that belongs to the seller.

The fix is not to learn every vendor's table by heart. It is to stop comparing credits at all, and to convert each quote into the only number that is the same across tools: what it costs you to end up with one ad you can actually run.

What a credit is, and what it is not

A prepaid token, metering compute

A credit is a prepaid unit of vendor compute. You buy a balance, the product spends it whenever a model runs, and the amount spent is decided by a schedule the vendor publishes. That schedule exists because the underlying costs really do differ. A still image at low resolution and a minute of 4K video with synced audio are not the same amount of machine time, and no single flat price covers both without either overcharging one or losing money on the other.

So credits are a reasonable internal accounting unit. The trouble starts when they are used as a public price. The number on the pricing card is denominated in a currency that has no external reference, no regulator, and no obligation to stay still.

The conversion table is the product

If you buy 2,000 credits, what you have actually bought is a claim on the conversion table. Run the same balance through Adobe's own published rates and watch it change size without anyone touching your account:

• At 1080p and 24fps, 100 credits per second, the balance is 20 seconds of video.
• At 720p and 24fps, 50 credits per second, the same balance is 40 seconds.
• At 540p and 24fps, 20 credits per second, it is 100 seconds.
• On a standard still image, 1 credit per generation, it is 2,000 presses.

The balance did not change. The table did. This is why a credit total on a pricing page is close to meaningless on its own, and why vendors are comfortable showing you a big one. Large numbers feel generous. Twenty seconds does not.

Why every vendor issues its own currency

Three reasons, and only one of them is about you:

• Repricing without a headline change. A vendor can move the cost of one feature without touching the plan price on the pricing card.
• One balance across many models. Per-model costs vary enormously, and a single currency genuinely does simplify the interface.
• Comparison friction. Cross-shopping requires arithmetic the pricing page never shows, and every extra step loses some shoppers.

You do not need to assume bad faith to act on this. Even a vendor with no interest in obscuring anything will publish a price you cannot compare, simply because it prices in its own units. The reader's job is the conversion.

Inside one vendor: nothing, 1 credit, and 20 credits

The zero-credit tier

Adobe's FAQ lists features that consume nothing at all. Under the heading for what does not use credits, the page states that "the following generative AI features do not currently consume generative credits," then names, among others:

• Adobe Express: Generate Text Effects, Generate Presentation and Clip Maker.
• Lightroom: Generative Remove.
• Photoshop: the Remove Tool.

Notice what that does to the mental model. These are real generative features, running real models, at a price of zero credits. A user who reasons "AI generation costs credits" has already been wrong once before reaching the video section.

The one-credit tier

Then the standard tier. The page reads: "Most standard generative AI features, like Generative Fill using Adobe Firefly models in Photoshop, use 1 credit per generation." That is the number most people carry in their heads as what a credit means, because it is the one that behaves like an intuition: one press, one credit.

It is also the tier that makes every other tier look strange. Once one press equals one credit, a feature that charges 20 for the same gesture reads as expensive, and a feature that charges 100 per second reads as a different product entirely.

The premium tier, where video lives

Adobe puts video in the expensive tier in its own words: "Premium generative AI features are advanced tools, like ones that can create videos, that need more power to run and thus require more generative credits per use." The rate card behind that sentence prices Generate Video at 100 credits per second at 1080p and 24 fps, 50 at 720p, and 20 at 540p, while Generative Extend in Premiere runs to 175 credits per second at 4K and 30 fps.

Within one vendor, then, the identical word costs nothing, 1 credit, or 20 credits depending on which button you press, and video is priced by the second on a separate card. If a single company needs three tiers and a rate table to explain its own currency, no cross-vendor credit comparison was ever going to hold.

Six meters wearing one word

Per generation, per second, per minute

The rate is only half of a price. The other half is the increment: the smallest slice the meter charges for. On the same Adobe page, credits are consumed per generation for standard features, per second for video, and per minute for Generate Soundtrack, which the table lists at 20 credits per minute. The increment can also differ between two sellers of the same model, which is why one avatar model bills in six-second blocks from the vendor that built it and by the second on a reseller: the lower rate is not always the lower invoice.

Those are three different shapes of bill from one vendor. A per-generation meter rewards long outputs and punishes iteration. A per-second meter does the reverse: it makes every extra second of a clip cost real money, and it makes a 15-second ad exactly three times the price of a 5-second one.

Per 1,000 characters, per object, per training

It gets more specific than that. Generate Speech is billed on input length, at "10 credits per 1000 characters with Firefly Speech." Rotate Object in Photoshop charges 20 credits per object, and the table adds a condition worth reading twice: "20 credits per object, only applied on first rotation for each." Firefly Custom Models (beta) charges 500 credits per training, plus 20 per generation afterward.

So on one page the meter counts, variously: presses, output seconds, output minutes, input characters, objects, and training runs. Six meters, one word.

Feature	What it charges	What the meter counts
Standard generation	1 credit	one press
Generate Video, 1080p 24fps	100 credits per second	output seconds
Generative Extend, 4K 30fps	175 credits per second	output seconds
Generate Soundtrack	20 credits per minute	output minutes
Generate Speech	10 credits per 1,000 characters	input characters
Rotate Object in Photoshop	20 credits per object	objects, first rotation only
Custom Models (beta)	500 credits per training	training runs
Express Clip Maker	no credits	nothing

Why the increment decides your bill

Two tools can advertise a similar per-second rate and still produce bills that differ by half. The rate is not what separates them:

• Per second against per clip. One meter scales linearly with length. The other rarely does.
• Per attempt against per accepted output. Discarded renders are either included in your price or invisible in it.
• Input against output. A voice feature billed on the script you typed is a different product from one billed on the audio it returned.

Before you compare rates, write down the increment beside each one. If a vendor does not publish it, that is your answer about how the bill will feel later. The same question applies to time as well as money, which is why we measured render times rather than quoting the marketing figure.

The conversion table moves under you

Adobe's own answer changed in three months

Credit schedules are dated documents, and they get edited. The same Adobe FAQ, in the revision stamped May 7, 2026, answered the question of whether all plans reach both feature types this way: "No, most Creative Cloud plans do not include access to premium generative features." In the revision stamped August 5, 2026, the answer to the same question reads: "Yes, most Creative Cloud plans include access to premium generative features."

Whatever you think of the change, notice what it does to a price you wrote down in June. The rate card did not have to move for your access to the expensive tier to move, and access is a price too when the tier you can reach decides whether video is available to you at all.

A quota is not a rate

Plan quotas invite the same confusion in the opposite direction. Adobe's plan table lists a premium-generation allowance per Firefly plan, each with unlimited standard generations on top:

• Firefly Standard: 2,000 credits for premium generations.
• Firefly Pro: 4,000.
• Firefly Pro Plus: 10,000.
• Firefly Premium: 50,000.

Those look like output numbers. They are not. Run the division against the video rate on the same page: 2,000 premium credits at 100 credits per second is 20 seconds of 1080p video per month. The quota is a budget denominated in a currency whose exchange rate lives two screens away, and only the division tells you what the month contains.

Write down the date you read it

Practical discipline, and it costs nothing. When you record a vendor's credit schedule, put four things in the same row:

• The rate, in the vendor's own units.
• The increment the meter charges in.
• The exact URL the rate came from, not the pricing page it was linked from.
• The date you read it. Anything older than a quarter is a lead, not a fact.

We apply the same rule to our own published numbers, which is why the figures further down are read out of the pricing configuration at the time of writing rather than copied from an older post. If a credit price is worth quoting to your finance team, it is worth a timestamp.

Convert everything to cost per finished ad

Here is the move that makes credit prices comparable. It has four steps, and only the first one is a judgment call.

Step 1: define one deliverable

Decide what "one finished thing" means for you, in output units, before you look at any price. For a performance marketer running paid social, a workable definition is:

• One vertical ad, roughly 15 seconds long.
• Three short shots, because a single unbroken take rarely holds attention.
• One product still for the end card or the thumbnail.

Be specific and keep the definition fixed across vendors. If you compare a 5-second silent clip against a finished 15-second ad with audio, the cheaper number will win and it will teach you nothing. Our own guide to how many ad creatives you need is a useful sanity check on how many of these you will actually want per month.

Step 2: find the rate and the increment

For each vendor, look up two things on the vendor's own page: the rate for the output you defined, and the increment it bills in. Then convert your deliverable into that vendor's units.

• Per-second vendors: multiply seconds by the per-second rate, at the resolution you would really ship.
• Per-clip vendors: look up the price of the exact duration you need, since these ladders are rarely linear.
• Per-generation vendors: count the presses, not the outputs, and include the ones that fail.

Step 3: divide the plan, not the credit

Now take the monthly plan price and divide it by the number of complete deliverables that plan produces. Not by the credit count. Not by the clip count. By the number of finished ads.

A worked pass on our own numbers, priced with the models Novoads runs and read from the pricing configuration rather than typed from memory. One display credit is 10 centi-credits internally, and the deliverable from step 1 breaks down like this:

• Three 5-second Seedance 2.0 clips at 720p: 3 credits each, so 9 credits.
• One product still through GPT Image 2 at medium quality: 0.3 credits.
• Total for one finished 15-second ad: 9.3 credits.

The entry monthly plan is $49 and carries 50 credits. Fifty divided by 9.3 is five complete ads a month, so the real number is $9.80 per finished ad, not the $0.98 per credit the arithmetic on the plan card would suggest. Both numbers are true. Only one of them is a price you can compare against a quote from another tool, or against what a human creator charges, which our breakdown of UGC creator rates puts in context.

Step 4: add the retries

The last step is the one everybody skips. Generative output is a lottery per render, so the honest denominator is finished ads you would actually run, not renders that completed successfully. If one in three attempts gets discarded for a warped hand or an off-brand read, your true cost per usable ad is half again what step 3 produced.

Measure it for a week rather than guessing, and keep the count crude:

• Renders started, including the ones you abandoned mid-queue.
• Ads actually shipped to a live campaign, not saved to a folder.
• The ratio between them, which is your real multiplier on every price above.

It is the single largest correction on this list, and it is the one no vendor's pricing page can give you. If you are building that habit, our notes on ad creative testing cover how to log the attempts as well as the winners.

How Novoads prices a finished ad

We price in credits too, so the honest thing is to publish the conversion rather than the balance. Every price below is computed from the same configuration the product charges against, which means a reprice moves this copy and your invoice on the same day rather than leaving a marketing number to rot.

What a clip costs

Every model below is selectable in Novoads today, and here is the part of the table a buyer actually needs, at 720p:

• Seedance 2.0, 5 seconds: 3 credits.
• Seedance 2.0, 10 seconds: 5 credits.
• Seedance 2.0, 15 seconds: 7 credits.
• Seedance 2.0 Mini, 5 seconds: 1.5 credits, the half-price variant.
• A product still through GPT Image 2: 0.3 credits.
• Google Veo 3.1: a flat 10 credits per video at any setting we offer.

You can see the same schedule model by model in our Seedance 2.0 pricing breakdown, and the underlying idea of prompting a model straight to video in our guide to text-to-video tools.

The increment is the clip, not the second

That ladder is worth reading as an increment, not just a list of prices. Under a strict per-second meter, a 15-second clip costs exactly three times a 5-second one. Under ours it costs 7 credits against 3, which is about 2.3 times. Longer output gets cheaper per second rather than staying flat, and that difference is invisible if you only compare headline rates.

The same logic runs the other way for short tests, so pick the increment that matches how you actually produce:

• A per-clip ladder suits you when you ship full-length ads, iterate on complete cuts, and want a predictable price per finished piece.
• A per-second meter suits you when you are cutting three-second hooks, testing openings in bulk, and would rather not pay a floor price for a fragment.

No increment is universally better. Knowing which side you are on before you buy a large balance is the whole point, and it is the question to carry into our roundup of tools that advertise no-cost tiers, where the headline rate is the least informative thing on the page.

What the plan buys

Fifty credits a month on the entry plan is sixteen 5-second Seedance 2.0 clips, or the five complete 15-second ads that the worked example produced. Subscription credits refresh each month and do not roll over; separately purchased top-ups are preserved. The entry point is a $1 trial for 3 days that converts to $49 a month, and the trial grant is deliberately small, enough to run the product end to end rather than to stock a library. You can start there and check our conversion table against the four steps above before you commit to anything.

The only currency that compares is the deliverable

Credits are an accounting convenience that became a marketing surface, and the drift is why two reasonable buyers can read the same two pricing pages and reach opposite conclusions. Nothing about that is fixable by a better credit. It is fixable by refusing to compare in credits at all.

Pick one deliverable, price it at every vendor in that vendor's own units, divide the plan by the ads it really produces, and add your retry rate. Four lines of arithmetic, and they turn a private currency back into a price. The vendor writes the exchange rate. You get to write the denominator, and the denominator is where the comparison actually lives.