Cloudinary pricing in 2026: what a credit actually buys for video

You have a 90-second 1080p clip, a 9:16 reframe to run and captions to burn in. Cloudinary's plan table prices itself in credits, and nothing on that page tells you how many credits that one clip costs.

Quick answer: Cloudinary pricing in 2026 runs on a single pooled credit. Cloudinary's published plan table lists Free at 25 credits and 3 users, Plus at $99/month ($89/month billed annually) for 225 credits, Advanced at $249/month ($224 annual) for 600, Advanced Extra at $549/month ($494 annual) for 1,350, plus a Pro pay-as-you-go option and custom Enterprise pricing. Cloudinary's billing documentation defines one credit as 1,000 transformations, or 1 GB of managed storage, or 1 GB of viewing bandwidth (2 GB of video bandwidth on paid self-service plans), or 500 SD / 250 HD seconds of video processing. For video, that means a single 90-second 1080p job draws on the same pool three separate times: once when it's processed, once as a stored derived copy, and once for every gigabyte delivered.

Most pricing explainers treat a credit as a prepaid unit of work, so a bigger plan buys more work. For images that's close enough to true. For video, a credit is an accounting unit that three meters draw from, and all three charge for the same file: processing, storage, delivery. That's why an account that ran a handful of transformations opens an invoice that's mostly storage and bandwidth.

These figures come off Cloudinary's pricing page and billing docs, read on November 1, 2026. Plan contents move, so check them before you commit a budget.

What a Cloudinary credit actually buys

A Cloudinary credit is one number you can spend four different ways, and the four exchange rates are published. Cloudinary's docs count video transformations per second of video processed, at a rate set by resolution: for progressive H.264, an SD second costs 2 transformations, an HD second 4, a 4K second 8, and anything above 4K costs 120.

Those rates match the credit table exactly. 1,000 ÷ 4 = 250 is the published HD video-second allowance, 1,000 ÷ 2 = 500 the SD figure.

A second of 1080p costs four transformations, so a 90-second 1080p job burns 360 of the 1,000 transformations inside one credit. Any "when do I outgrow the free tier" math that treats a transformation as a unit of one is wrong by orders of magnitude for video.

Above 4K the rate goes vertical. At 120 transformations per second, a 10-second 8K clip costs 1,200 transformations, more than a full credit for ten seconds of footage.

One 90-second clip, through all three meters

Take a real job instead of a rate card: a 90-second 1080p upload, reframed to 1080x1920, captions burned in, then delivered 100 times. Assume the source lands at about 8 Mbps (90 MB) and the vertical output at about 6 Mbps (68 MB). On Plus, $99 for 225 credits works out to $0.44 per credit.

At $0.44 per credit, that clip costs roughly $1.72, and the transform you asked for is 9% of it. Delivery is 87%. Skip the views and pull the output back to your own bucket once, and the same job lands at about 0.55 credits, or $0.24.

Two details make that table behave worse than it looks. Cloudinary's storage usage includes a cached copy of every derived asset generated by a delivered transformation, so each platform variant you render becomes its own stored file. Render 9:16, 1:1 and 4:5 off one 90-second master and you've spent 1,080 transformations and store four copies. The docs also note derived assets count against your storage quota but aren't backed up, so you pay rent on files with no durability guarantee.

Regeneration is the second detail. A transformation is counted when the derived asset is generated, eagerly at the API call or on first delivery request, and repeat hits on the same URL are free. Invalidate and regenerate that asset, say after editing a named transformation, and the usage is charged again even though an identical file existed a minute earlier. One template tweak across a 500-clip library re-bills the whole library.

Why the invoice doesn't reset when you expect

On self-service plans, Cloudinary's support docs describe transformations and bandwidth as measured over a rolling 30-day window rather than resetting at the start of a calendar month, with storage reflecting your current total. Pro pay-as-you-go and Enterprise count per billing month instead.

Fire off a bad batch of 400 reframes on the 3rd and those credits don't fall off on the 1st. They sit in your usage for the next 30 days, exactly when you'd be running the month's real work.

Past 100% of quota on a fixed plan, Cloudinary documents Console warnings, emails to admins at around 90%, prompts to upgrade or cut usage, and possible account disabling if usage stays over. On Pro pay-as-you-go, over-quota usage is billed as overage on the normal billing date rather than blocked.

Two more lines sit outside the credit pool entirely. Add-ons are billed separately with their own pricing tiers and don't consume plan credits, so auto-transcription for captions isn't in the 225. And the Free plan caps video uploads at 100 MB, which a 90-second 1080p phone capture can exceed.

Cloudinary's docs don't say whether a failed video transformation still consumes quota.

What $99 honestly buys

Four credits for one captioned vertical clip sounds expensive, and in isolation it is. The $99 covers four products though: a digital asset manager with user roles, an image transformation pipeline, a video player, and a CDN that delivers all of it. Each would cost money separately, so the price is fair for anyone using all four.

The mismatch shows up when your pipeline needs exactly one of them. If the job is "reframe this, burn captions, hand the file back, I already have S3 and Cloudflare," you're buying a DAM, a player and a CDN to get a transcode, and the pooled credit bills you for a hosting relationship you didn't want. Same shape as Mux's three meters, where encoding is priced at $0 because delivery is where the model lives.

The processing-only invoice, side by side

A month of 100 videos, priced both ways: 90 seconds each at 1080p, reframed to 9:16 with captions burned in, outputs pulled back to storage you already pay for.

FFmpeg Micro's meter is published in full and has one dial: 1 token per 10 seconds of input, metered per second with no rounding up to the minute. Status checks, job listings and downloads are free, and failed jobs are never billed. A new account gets 200 tokens at sign-up with no card, about 33 minutes of input, or roughly 11 complete reframe-plus-caption jobs on 90-second clips. Starter is $19.99 for 2,500 tokens a month, Pro is $49.99 for 7,000, and top-ups work on any plan including Free at 500 tokens for $5, 1,200 for $10 or 3,000 for $25. Purchased tokens don't expire while the account is open. Nothing is stored and nothing is delivered, so that 1,800-token month costs 1,800 tokens again next month and never grows a tail. You can run it from code, from n8n, Make and Zapier, or from an AI agent over MCP, and per-second metering means trimming 30 seconds out of an hour-long recording bills 30 seconds.

Be careful with third-party pages quoting FFmpeg Micro's numbers. Several still list a $19 Starter with a per-minute rate, a $89 Pro tier, a "Scale" plan, and billing rounded up to the minute. None of that is live, and the rounding claim inverts the billing rule that matters here.

Which meter fits your pipeline

Pick your meter by what happens to the file after the job finishes. If the output lives on Cloudinary and gets delivered from Cloudinary, a pooled credit is the right currency and the plan table is the real price. If the output leaves immediately for a bucket and a CDN you already run, every credit you spend on storage and bandwidth is a tax on a product you're not using.

A pooled media platform is the better call when you want URL-based on-the-fly transforms your front end can request at render time, a player, signed delivery, image and video in one asset library, and a team that browses assets in a UI. A per-second processing API can't do any of that. FFmpeg Micro is an API, not an editor and not a host.

Per-second processing wins when the work is mechanical and finite: reframes, caption burns, watermarks, audio extraction, thumbnails, format normalization for an AI model's input. It's the same split as pricing machine work separately from human work. And if your actual job is converting a 4K master into 1080x1920 for TikTok, check the 4K rate first: 90 seconds of 4K input is 720 transformations at 8 per second, double the 1080p figure, for a file you're about to downscale.

FAQ

How many Cloudinary credits does a one-minute 1080p video cost?

Processing one minute of 1080p video costs 240 transformations at Cloudinary's published rate (60 seconds at 4 transformations per second), which is 0.24 credits, or about 11 cents at the $0.44-per-credit Plus rate. That covers processing only. Storing the original plus each derived copy and delivering them draws separately on the same pool, and on a real workload delivery is usually the biggest line.

Is the Cloudinary free tier enough if I only process video?

The 25 free credits cover 6,250 HD video-processing seconds, about 104 minutes of 1080p, if literally nothing else touches the pool. In practice you can't, because Cloudinary caches a copy of every derived asset against your storage quota and you spend bandwidth pulling the output back out. The Free plan also caps video uploads at 100 MB per file.

If a Cloudinary transformation fails, do I still lose the credits?

Cloudinary's documentation doesn't state either way, so confirm it with their support before you build retries on top of it. FFmpeg Micro publishes the opposite rule explicitly: failed jobs are never billed, which is what makes a retry loop safe to automate.

That $1.72 for one captioned vertical clip is only 9% processing, and the rest is rent on storage and delivery you may already pay for elsewhere. If the reframes and caption burns are the part you want off your plate, run the arithmetic on your own footage at FFmpeg Micro's pricing comparison, then Sign up free and spend the 200 starter tokens on tonight's batch. No FFmpeg to install, no server to keep alive, no card.