What AI media generation actually costs (real numbers)
6 min read
Real prices for AI images, video, voiceover and music, how per-output credits work, and how to budget a content pipeline without lock-in.
Budgeting AI media is confusing mostly because pricing models differ so much between tools. Some charge flat subscriptions with monthly caps, some sell credits that expire, some hide the per-generation number entirely. Before committing a content pipeline to any tool, you want one question answered plainly: what does one output cost, in dollars?
The numbers below use Dolly's pricing as the worked example, since every price is public on the pricing page.
The unit: one credit is one cent
Everything is priced in credits at a face value of one cent each. Every generation shows its exact credit cost before you run it, and the same number is what gets charged. There are no per-seat fees and no meter running in the background. Failed or timed-out generations refund automatically, which adds up to real money once you generate at volume.
What outputs actually cost
- Images: 4 to 20 credits each. The fast draft tier starts at 4 credits, most of the lineup sits between 5 and 15, and only premium models at their highest resolutions reach 20. The default flagship, Flux 2 Pro, is 5 at standard resolution, so a hundred product-shot variants on it is about five dollars.
- Video: roughly 15 to 250 credits for a short clip at default settings. The cheapest draft models start around 15 credits, the mid-size Seedance Mini is 86, and the default flagship is 250 for five seconds at 720p; longer durations and 1080p or 4K renders run several times these numbers. Video is where budgets live or die, which is why drafting cheap and finishing expensive (the ladder in the product video guide) is the standard play.
- Voiceover: 4 to 15 credits per 1,000 characters. Budget voices start at 4 credits and the ElevenLabs tiers run 8 to 15. A 60 second script is roughly 800 characters, so narrating a video costs about a dime.
- Music: 12 credits per generation, two tracks included. That is the Suno models, which go up to six minutes per track. Eleven Music instead prices per started minute at 125 credits, returning one polished take at the exact length you set. A Suno soundbed is pocket change, and even the longest Eleven Music track runs about six dollars.
A complete short ad (four video drafts, two finals, narration, music) lands around nine dollars. A month of daily social images with heavy variant testing might be twenty. These are real production budgets, not enterprise line items.
Credits vs. subscriptions
The honest comparison is not about any single competitor, it is about pricing shapes. Subscription caps punish lumpy usage: founders generate in bursts before a launch and then quietly pay for two idle months. Expiring credits are the same problem wearing a different hat. Per-output credits that persist fit bursty work better, and you can see the fit in the numbers above: your cost tracks what you shipped, not what you subscribed to.
Membership still earns its keep as a plain discount. On Dolly, plans from $9 to $349 a month buy the same credits at a better rate (down to about $0.0084 per credit against the one cent face value), everything is included on every plan, and unused credits roll over rather than vanishing at midnight (monthly grants pause only while you are sitting on more than three months of your plan's credits). One-time packs cover the no-subscription case. The full ladder is on the plans page.
Budgeting a pipeline: three rules
- Draft on cheap models, finish on flagships. The 5 credit image and the 86 credit clip exist to absorb your iteration. Spend big only on keepers.
- Count the variants behind each keeper. If you keep one image in five, your real image cost is five times the sticker. Budget for the ratio you actually observe, and consider batching the variants so triage stays fast.
- Let the button do the accounting. Because the price sits on the generate button, drift is visible at decision time. A team that sees "250 cr" before every flagship clip self-corrects; one that discovers costs on a monthly bill does not.
The entry plan runs the image and voice numbers above many times over — see pricing for the current figures.