Plan → Budget → Prompt
Claude Video
Cost Calculator
Plan the film before the bill. Turn a 30-second motion-video brief into an editable token and production budget.
Estimate your videoFree · no sign-in · no generation charges
What are you making?
Illustrative token recipes, not measured video averages. Replace the assumptions with your own observations.
Planned model calls · 6 steps
| Step | Input tokens | Output tokens | USD |
|---|---|---|---|
| Script | 2,800 | 600 | $0.0232 |
| Storyboard | 3,400 | 960 | $0.0328 |
| Animation code | 4,360 | 3,600 | $0.08944 |
| First review | 7,960 | 400 | $0.03984 |
| One-shot repair 1 | 8,360 | 600 | $0.04544 |
| Review after repair 1 | 8,960 | 400 | $0.04384 |
A video plan is more than one prompt.
Count the whole making process
The estimate includes a script, a per-shot storyboard, animation code and a model review. Each revision adds a repair plus another review. Reworking the whole film produces more code than changing one shot.
Every call includes your source-context and overhead assumptions, plus modeled preceding outputs. Overhead is the total allowance for all remaining input on each call, including prior user messages and tool results. Retaining outputs makes input grow. Discarding an attempt does not erase the tokens already billed.
Use scenarios, then check actual usage
The three scenarios scale assumed output and recalculate the later input context. They are illustrations, not statistical confidence intervals, benchmarks or a promise that the high scenario caps your bill. Each modeled request must stay within Opus 5.5’s 1M context and 128K output limits. See the official model specifications ↗.
Use Anthropic token counting ↗ to estimate a real request’s input. Actual output, thinking, tool loops and repairs are only known after calls complete. Use the Claude Opus 5.5 cost calculator when you have actual usage and cache counters.
Official Standard rates: $4/million uncached input tokens and $20/million output tokens. Anthropic pricing ↗. The recipe’s token quantities are our editable assumptions; the unit rates are official.
Before you spend
Does Claude charge per finished video minute?
Claude API work is priced by billed tokens, not finished video seconds. A 30-second film can require very different amounts of code, source material and revisions. This first version budgets the 30-second HTML/CSS motion-film workflow, not Sora/Veo/Runway clips or the separate 9:16 Short generator.
Are these token recipes real measured averages?
No. They are transparent, editable planning assumptions. Changing the type or complexity loads a different illustrative output assumption. Use your own completed-run usage to calibrate the recipe; no paid benchmark is claimed.
Can the final cost exceed the high-output scenario?
Yes. Unexpected tool loops, failed attempts, larger source material and additional revisions can all increase spend. A scenario is not a provider spending limit.
Why are images, voice and rendering separate?
Image providers, voice services and rendering compute use their own prices and units. Their USD quotes belong in the production budget, not the Claude token total. A blank quote stays unpriced, and local compute is not automatically zero-cost.
Is this my Claude Pro or Max bill?
No. This is first-party API list-price planning. Subscription allowances, usage limits and extra-usage billing are separate. The API rates page explains those differences; nothing here changes your plan or calls a provider.
What happens after I build the prompt?
The Prompt builder receives this video type, shot count, motion complexity and revision plan. You supply real product facts and get a matching storyboard and prompt for your own Claude Code workspace. The existing starter renders a silent 30-second 16:9 MP4 after authorized local commands; this tool does not generate or publish a video.