The batch video API turns one-off generation into a production line: you submit a manifest of dozens or hundreds of jobs, the queue works through them, and you pull back finished, quality-gated cuts when they are ready. Instead of babysitting a single render in a browser tab, you describe the whole run once in code and walk away. For any team that needs creative at real volume, a localization pass, a thousand product variants, a week of social, that is the difference between shipping and stalling.
What a batch video API is for
Manual generation is fine when you need one film, it collapses the moment you need fifty. You end up opening tab after tab, kicking off renders by hand, and tracking which ones finished, which failed, and which still need a fix. A batch API removes that ceremony entirely. You hand the platform a list of jobs and a place to put the results, and it owns the rest: queueing, claiming, generating, scoring, rendering, and uploading.
The jobs that benefit most share a shape. They are repetitive but parameterized, the same creative recipe with different inputs. A few concrete examples:
- Product catalogs. One shot board, swapped product name, image, and price across every SKU.
- Localization. One master cut, re-voiced and re-captioned in a dozen languages.
- Ad variants. One concept, ten hooks, three aspect ratios, for a structured creative test.
- Programmatic content. A new explainer for every row in a spreadsheet of topics.
If you are still wiring up your first integration, start with the CoreReflex API quickstart to get authenticated and make a single generate call, then graduate to batch once that one job works end to end.
How a batch run works, step by step
A batch run is a queue, not a single blocking request. You submit, the platform accepts the work, and you collect results asynchronously. The flow looks like this:
- Authenticate with a scoped API key. Mint a key for your backend service or CI job. The same key drives single generations and batch submissions.
- Build a manifest. Express each job as a structured entry over the programmable JSON engine, a brief, a shot board, or a template fill, plus any per-job parameters. Planning and scoring are free, so you can construct and validate the manifest without spending anything.
- Submit the batch. Post the manifest as one request. The platform returns a batch ID and accepts every job into the queue. Generation is the only step that consumes credits, and it only starts now.
- Let the queue work. Each job is claimed, generated shot by shot, scored against the quality gate, and, on the render path, claimed by a render-worker, rendered, and uploaded to your storage as an asset row. The deterministic render path means the same manifest produces the same cut every time.
- Pull finished results. Poll the batch status or subscribe to completion, then read back each job's assets, scores, and trace. Successful jobs are ready to ship; failures come back with a reason rather than silence.
The key mental shift is that submission and completion are decoupled. Your code does not hold a connection open for an hour, it fires the manifest, then checks back.
Anatomy of a batch manifest
A manifest is just structured data, which is what makes it diffable, versionable, and reproducible. A single job entry typically carries:
| Field | What it controls |
|---|---|
brief / board | The creative input, a one-line brief the Director boards, or an explicit shot board |
params | Per-job variables: product name, language, hook copy, aspect ratio |
brand | The brand kit to enforce, colors, fonts, voice |
output | Format and encoder tier for the render |
budget | A per-job or per-batch credit ceiling |
webhook | Where to notify your service when the job completes |
Because the manifest is data, you generate it from your own source of truth. A loop over a product feed, a translation table, or a CSV becomes the manifest. That is the same principle behind generating a film from a single sentence, you describe intent and let the Director board the shots, applied across an entire list at once.
The quality gate runs on every job in the batch
Scale is worthless if it lowers the bar. It does not here. Every shot in every job runs through the same quality gate as an interactive generation: each shot is scored on prompt match, sharpness, motion coherence, on-screen text legibility, on-brand fit, and claims risk. A shot that clears the bar moves on. A shot that fails triggers selective regeneration, the Director re-rolls just that shot, carrying forward the plan, the camera move, and the continuity anchor, and keeps the shots that already passed. You do not get a pile of clips to inspect by hand; you get cuts that already cleared the checks.
If you want to read those scores programmatically, to set your own pass threshold, route borderline jobs for human review, or block a deliverable that did not meet spec, the Eval API lets you score outputs and act on the result inside your own pipeline.
Provenance and budgets for the whole run
Every generation in a batch rides a portable trace: which model produced the shot, with which prompt and parameters, and the score it earned. So a batch of five hundred assets is not five hundred mysteries, each one is reproducible and auditable, which matters when a client or a compliance reviewer asks how a specific clip was made.
Cost stays under your control because of how credits work. Planning and scoring across the entire manifest cost nothing, so you can validate a thousand-job run before a single credit is spent. Only generation draws down credits, and you can cap spend per job or per batch with a budget ceiling. For runs you want to fire on a schedule rather than on demand, a nightly product refresh, a weekly social drop, pair the batch API with Autopilot to schedule hands-off generation so the manifest builds and submits itself. See the current rates on the pricing page.
Where batch fits in the platform
The batch API is one surface over the same engine that powers everything else, so it composes cleanly with the rest of the platform. Ground every job in your own facts with the knowledge base and RAG layer, cut latency and token cost on repetitive briefs with the context cache, and drive the whole thing from an agent through the MCP server. The full reference for keys, endpoints, and the SDK lives in the developer docs, and you can browse the rest of our developer guides for deeper dives on each piece. The throughline is consistency: the same Director loop, the same quality gate, and the same deterministic render path whether you generate one cut by hand or ten thousand from a manifest.
Frequently asked questions
How do I generate many videos at once?
Submit a batch manifest, a structured list where each entry is a job with its own brief or board and parameters, in a single API call. The platform queues every job, generates and scores each one, renders the passes, and uploads finished assets to your storage. You poll the batch status or receive a webhook when jobs complete, then pull the results.
Does batch generation still run the quality gate?
Yes. Every shot in every batched job runs through the identical quality gate as an interactive generation, scored on prompt match, sharpness, motion coherence, text legibility, on-brand fit, and claims risk. Failed shots trigger selective regeneration rather than restarting the job, so volume never means a lower bar.
What happens if one job in the batch fails?
Failures are isolated. A job that cannot pass after regeneration comes back flagged with a reason, and the render path's retry and dead-letter queue handle transient errors automatically. The rest of the batch finishes normally, so one bad job never blocks the run.
Can I cap how much a batch run spends?
Yes. Because planning and scoring are free and only generation consumes credits, you can build and validate the entire manifest at no cost. You then set a per-job or per-batch budget ceiling so a run stops before exceeding the limit you define.
Generate at scale, without lowering the bar
The batch video API is how you move from making one video to running a creative supply line, a manifest in, finished and quality-gated cuts out, with a trace on every asset and a budget you control. It is the same agentic Director and the same deterministic render path you already trust, pointed at a list instead of a single job.
Describe your first film and watch the loop run. It is free to start, no credit card. Start building on CoreReflex.