Business & product
AI writes the JSON, Zvid renders it.
This is not text-to-video. An AI agent authors a structured JSON project through the MCP server, and Zvid renders it deterministically with FFmpeg — so the same document always produces the same frame-accurate result.
Agent-authored
AI plans the video, FFmpeg renders it
"AI-generated video" with Zvid does not mean diffusion or generated pixels. It means an AI agent — Claude Code, Cursor, Codex, or any LLM — authors a JSON project describing scenes, elements, transitions, and captions, then Zvid renders that document with FFmpeg and returns a hosted cdn.zvid.io URL. Because the engine is deterministic, the same JSON always produces the same frames.
The agent connects through Zvid's MCP server, which exposes 30 tools plus schema-aware authoring so the model builds valid projects instead of guessing at shape. Before a single credit is spent, a free pre-flight validator resolves templates and applies your plan limits, and a repair tool fixes malformed projects. You get output an agent and your brand can both rely on.
What you get
What agent-authored video gives you
30 tools for agents
Zvid's MCP server exposes 30 tools to AI agents like Claude Code, Claude Desktop, Codex, and Cursor, so the model authors valid JSON projects instead of freehanding structure.
Free pre-flight validation
The validator resolves templates and applies your plan limits without spending any credits, so an agent knows a project will render before you commit to it.
Deterministic FFmpeg output
Rendering runs on FFmpeg, not a headless browser, so the same JSON produces the same frame-accurate result every time — reproducible enough to trust in production.
Self-repairing projects
A dedicated repair tool fixes malformed JSON, and best-effort bulk validation returns invalid items in itemErrors by index while the valid ones keep rendering.
Schema-aware authoring
The MCP integration is schema-aware, guiding the agent to the correct field shapes across scenes, elements, transitions, and 12 caption modes as it builds the document.
Get pinged on completion
HMAC-SHA256 signed webhooks fire render.completed and render.failed with 5 retries, so an agent-driven pipeline reacts the moment a render lands.
30
MCP agent tools
FREE
Pre-flight validator
FFmpeg
Deterministic engine
1,200 free
Credits to start
Under the hood
The features that make it work
MCP Server & AI Agents
A Model Context Protocol server exposes 30 Zvid tools to any AI agent — with schema-aware authoring and a free validation step so generated JSON renders first try.
Templates, Variables & Logic
Save a project once, render it with different data forever. Placeholders, conditional scenes, and array iteration resolve server-side before validation.
JSON Video API
Render a finished, hosted video from one authenticated POST. Send an inline JSON project or a template id and get back a job you can poll or get pinged about.
Subtitles & Captions
Word-timed captions with 12 animated styles — karaoke, highlight, typewriter, pop, and more. Import SRT, VTT, ASS, or Whisper JSON, or author by hand.
Who it's for
Teams that build this with Zvid
AI engineers & agent builders
Your agent authors the JSON; Zvid renders it deterministically with FFmpeg — video or images, never text-to-video. The MCP server ships 30 tools plus a free pre-flight validator, so you catch errors before a credit is spent.
Explore →Developers
Make rendering an HTTP call. One authenticated POST returns a 202 with a job id, and a per-request webhook pings you when your hosted cdn.zvid.io URL is ready — no polling, no headless browser.
Explore →No-code & automation builders
Wire Zvid into your automations with no backend. Trigger renders from a spreadsheet, CRM, or form via n8n or Make, get pinged on completion, and fan out with a single bulk call.
Explore →Related use cases
More for teams like yours
Personalized video at scale
One video per lead, customer, or account — your rows merged into a template and rendered at scale. POST up to 500 renders in a single call, each pulling {{variables}} from one row, and get back hosted cdn.zvid.io URLs.
Explore →Faceless & captioned video
Build no-camera short-form the API way: stock media plus your own voice-over, Jamendo music, and TikTok-grade animated captions in 12 modes, rendered by FFmpeg as one hosted video.
Explore →Data-driven & report videos
Turn a metrics feed into per-account report videos. Bind your numbers to a template, fan it out with one bulk call, and get back hosted MP4s or images — rendered by FFmpeg, so every figure lands frame-accurate.
Explore →FAQ
Frequently asked questions
Does Zvid generate video from a text prompt? +
No. Zvid does not do text-to-video, diffusion, or generated pixels. An AI agent authors a structured JSON project — typically through the MCP server — and Zvid renders that JSON deterministically with FFmpeg.
How does an AI agent connect to Zvid? +
Through the MCP server, which exposes 30 tools with schema-aware authoring to agents like Claude Code, Cursor, and Codex. A free pre-flight validator lets the agent confirm a project is valid before any credits are spent.
Why does deterministic rendering matter for AI output? +
Because the engine is FFmpeg rather than a headless browser, the same JSON always renders the same frames. An agent can regenerate a document and trust the video matches, which matters when a model produces content at scale for your brand.
Explore more: see every use cases
Ready to ship video at scale?
Start free — your first render is only a JSON payload away.