> ## Documentation Index > Fetch the complete documentation index at: https://docs.lavendly.ai/llms.txt > Use this file to discover all available pages before exploring further. # Credits & plans > How the ledger works, what each plan gives you, and why agents should reserve before they render. ## What costs credits **Generation is the only paid step.** Rendering - stitching your generated clips into the final video - is **free**, and it's reuse-aware: it folds in clips you already generated and never re-charges them. So "generate the scenes, then render the video" costs exactly the scenes, and re-rendering to tweak the edit or audio is free. Use `estimate_cost` to see the price before you spend. ## How limits actually work Every paid generation goes through a server-side credit gate: 1. The API estimates the action's credit cost from the model + duration (or text length, for voice). 2. It reserves that many credits from your balance. Reserve fails → immediate `402 insufficient_credits` with the exact `needed` / `have` you can show in your UI. 3. The work runs and the hold is **settled to what was actually delivered**: every clip we produced is charged (it's a clip the provider billed us for), and any clip that never generated is released. That means **you only pay for clips actually produced**. A render that fails after making some clips charges only for those - you keep them and re-render for free - and one that produces nothing costs nothing. A timeout, a model outage, or a content-policy rejection on a clip never charges for that clip. ### What hitting the limit looks like ```json theme={null} HTTP/1.1 402 Payment Required { "error": { "code": "insufficient_credits", "message": "Need 10 credits, you have 4.", "needed": 10, "have": 4, "plan": "starter", "topup_url": "/Pricing" } } ``` The SDK and the canvas both surface a "Top up" CTA on this code, clicking it sends the user straight to Pricing. ### Per-plan ceiling | Plan | Monthly credits | Hits zero? | | ------- | --------------: | ------------------------------------- | | Free | 45 | Top up or wait for the monthly refill | | Starter | 200 | Top up or upgrade | | Creator | 450 | Top up or upgrade | | Studio | 1200 | Top up or upgrade | Unspent credits **carry over** every month, the monthly refill tops you back up to the plan ceiling, it doesn't reset you to zero. ### Allowlisted accounts Internal accounts on a server-side allowlist bypass the credit gate entirely. The monthly USD cap still applies, it's not a free pass to infinite spend. ## The reserve / settle pattern Every paid action holds credits first, then settles to what was delivered: ``` reserve(amount) ───────────▶ token + hold placed │ settle(token, delivered/reserved) │ ├── all delivered ▶ full amount debited ├── some delivered ▶ charge that fraction, release the rest └── none delivered ▶ full release (free) ``` Why hold first? Because a multi-scene generation can fail mid-flight. Reserving up front ensures you can't oversell credits even if 50 renders fire in parallel and one hits a model timeout 90 seconds in. Settling to the *delivered* fraction means a partial failure charges only for the clips that were actually produced - never a blanket refund of work the provider already billed us for, and never a charge for clips you never got. The SDK does this for you transparently. Agents talking to the API directly just reserve before generating; the server settles the hold when the job reaches a terminal state. ## Plans | Plan | Monthly credits | Price | Max resolution | Notes | | ------- | --------------: | ----: | -------------- | ----------------- | | Free | 45 | \$0 | 720p | Onboarding | | Starter | 200 | \$19 | 720p | Personal projects | | Creator | 450 | \$39 | 1080p | Most popular | | Studio | 1200 | \$99 | 4K | Teams | Unspent credits **carry over** month to month. The "monthly pool" just refills up to the plan cap, it doesn't reset to zero. ## Per-action cost Costs are denominated in credits, not dollars, so we can adjust upstream pricing without you re-budgeting. Roughly: * 720p video, 5 s: **5-10 credits** depending on the model picked * Voiceover, 30 s: **2 credits** * Music bed, 30 s: **3 credits** * Still image: **1 credit** The authoritative per-model numbers come from `get_status`, your agent should read those before committing a multi-shot plan. ## Monthly cost cap Independent of credits, every account has a **monthly upstream cost ceiling**. When that ceiling is hit, paid endpoints return `402 monthly_cost_cap` until the next month rolls over. This is a safety belt for autonomous agents, even an unlimited plan won't let a runaway loop burn an arbitrary amount of money in a weekend. See [`GET /v1/usage/monthly`](/api-reference/ledger/usage) and the [cost-aware skill](/agent-skills/cost-aware). ## Per-video limits Independent of credits and the monthly cap, every single generation is bounded by two hard guardrails so a prompt - or a runaway agent - can't queue an absurdly long or expensive video: | Limit | Default | Env var | | --------------------------------- | ------------: | -------------------------- | | Max final-video duration | 3 min (180 s) | `MAX_VIDEO_DURATION_SEC` | | Max upstream spend per generation | \$30 | `MAX_VIDEO_GENERATION_USD` | The duration cap is the **sum of all scene durations**; the budget cap is the estimated upstream cost of the clips actually being generated (reused clips cost nothing). `create_render`, `generate_scene`, and the single-clip video endpoint all enforce them **before** queuing, returning `400 video_too_long` or `402 video_over_budget` with an actionable message. Both are operator-tunable via the env vars above - no code change needed to raise or lower them. Call `estimate_cost` first to see where a workflow sits against these limits before you spend.