Quickstart (Beta)

Beta. The Video API is in beta. Everything below runs against production.

1. Get your API key

Create a key in the dashboard under Settings > API Keys. Every request authenticates with the Authorization: Bearer YOUR_API_KEY header.

2. Provide your input video

input_url (and input_urls) accept a publicly downloadable https URL to a media file — we stage your media into eachlabs storage automatically before processing. Private footage: a presigned URL from your own storage works — any https URL the engine can fetch is staged the same way (see Acceptable Use). s3:// input URIs from your onboarding keep working unchanged, and you can chain jobs by passing a previous job’s output URL as the next job’s input. Local files: upload them to eachlabs storage and pass the returned public_url directly as input_url.

The URL must point directly at a media file, identified by its file extension. The staged extensions are: video .mp4 .webm .mov .avi .mkv .flv; audio .mp3 .wav .ogg .flac .m4a .aac .opus .wma; image .png .jpg .jpeg .gif .webp .bmp .tiff; subtitle .vtt .srt .ass; color LUT .cube. A URL we cannot stage (unreachable, not public, or not identifiable as media) is rejected at $0 with PROVIDER_REJECTED and the offending URL in the message.A URL with no path extension at all can declare its type with a mime_type= query parameter (the parameter is only consulted when the URL path carries no extension — it cannot override an unrecognized extension). The value is the MIME type with the separators encoded as underscores — the first _ becomes /, any later _ becomes -: mime_type=video_mp4 → video/mp4, video_quicktime → video/quicktime, audio_x_wav → audio/x-wav. Any video/*, audio/*, or image/* type is accepted (the subtype is not checked against a list); non-media types are rejected. Note .ts artifacts (hls_ladder segments, concat_copy container: "ts") are terminal today: .ts is not a staged extension and the extension blocks the mime_type= fallback, so they cannot be chained as inputs.

.cube LUTs stage only when the URL path carries the .cube extension — the mime_type= fallback accepts media types only, so an extensionless LUT URL is not staged. sendcmd scripts are not staged from public URLs at all: host them on eachlabs storage instead — an s3:// URI from your onboarding or a cdn-us.eachlabs.ai URL.

3. Submit a job

Any capability from the reference works the same way — a request carries capability, input_url, and params.

curl -X POST "https://api.eachlabs.ai/v1/prediction" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "eachlabs-video-api",
    "input": {
      "capability": "transcode",
      "input_url": "YOUR_INPUT_URL",
      "params": { "vcodec": "h264", "crf": 28 }
    }
  }'

import requests

resp = requests.post(
    "https://api.eachlabs.ai/v1/prediction",
    headers={"Authorization": "Bearer YOUR_API_KEY"},
    json={
        "model": "eachlabs-video-api",
        "input": {
            "capability": "transcode",
            "input_url": "YOUR_INPUT_URL",
            "params": {"vcodec": "h264", "crf": 28},
        },
    },
)
prediction_id = resp.json()["predictionID"]

const resp = await fetch("https://api.eachlabs.ai/v1/prediction", {
  method: "POST",
  headers: {
    "Content-Type": "application/json",
    "Authorization": "Bearer YOUR_API_KEY",
  },
  body: JSON.stringify({
    model: "eachlabs-video-api",
    input: {
      capability: "transcode",
      input_url: "YOUR_INPUT_URL",
      params: { vcodec: "h264", crf: 28 },
    },
  }),
});
const { predictionID } = await resp.json();

A successful submit returns:

{ "status": "success", "message": "Prediction created successfully", "predictionID": "d63b9b70-..." }

4. Poll until terminal

curl "https://api.eachlabs.ai/v1/prediction/PREDICTION_ID" \
  -H "Authorization: Bearer YOUR_API_KEY"

The typical flow is created → processing → success. The complete status set is created, processing, success, error, cancelled — a failed prediction reports error, never "failed". created is the first status you see; the API never returns starting, so do not branch on it. Poll every 3-5 seconds: the platform’s own status checks are quantized to ~5 s, so tighter loops buy nothing. Most short jobs finish in well under a minute (a 100-job burst of short transcodes drains in ~15-30 seconds). A prediction sitting in created unusually long is queued behind a concurrency cap — keep polling; see Rate limits & concurrency.

{
  "id": "1f0fdf4d-...",
  "input": {
    "capability": "transcode",
    "input_url": "https://your-host.example/source.mp4",
    "params": { "vcodec": "h264", "crf": 28 }
  },
  "logs": null,
  "status": "success",
  "output": ["https://cdn-us.eachlabs.ai/uploads/9b967c93-....mp4"],
  "metrics": { "predict_time": 194.86, "cost": 0.2865, "tier": "cpu", "billed_seconds": 191, "output_gb": 0.0024 },
  "urls": {
    "get": "https://api.eachlabs.ai/v1/prediction/1f0fdf4d-...",
    "cancel": "https://api.eachlabs.ai/v1/prediction/1f0fdf4d-.../cancel"
  }
}

output — hosted artifact URL(s) for media capabilities, or a structured JSON object for the analysis capabilities (probe, scene_detect, silence_detect, audio_analysis). The exact JSON per output class is in Response shapes.
metrics.cost — what the job actually billed, in USD: billed_seconds × $0.0015, plus delivery on `output_gb` beyond the 200 MB free allowance per job ($ 0.30/GB on the excess) — see the formula. Rejected, failed, and cancelled jobs bill 0 (see Failed compute for how failures are metered).
metrics.billed_seconds, metrics.output_gb, metrics.tier — the billed quantities, present on every successful prediction. Failed predictions carry only cost and predict_time.
metrics.predict_time — the prediction’s full wall time in seconds, from processing start to completion detection, including internal queueing and poll quantization. It is not the billed quantity: cost comes from metrics.billed_seconds × $0.0015. In the example above `billed_seconds` reads 191 → `191 ×$ 0.0015 = $0.2865, while predict_time` reads 194.86 — see the formula.
input — your submitted input, echoed back. logs is present on every poll and is null today.
urls.get — this prediction’s poll URL. urls.cancel — its cancel endpoint (POST, same Authorization header); cancellation is record-only and a cancelled prediction never bills — see Cancelling a prediction.
On status: "error" the output object carries error_code and error_message — except on a dispatch failure at submit, which writes a minimal envelope carrying only error, so guard for their absence before reading them. The field-by-field envelope with two real captured examples, and the billing impact per class, are in the error catalog.
Download or persist artifacts promptly; treat artifact URLs as valid for at least 24 hours.

5. Optional: completion webhooks

Instead of polling, pass the platform’s standard webhook_url when creating the prediction and receive a callback on completion — see the webhooks documentation.

Next steps

Capabilities Reference

Every capability with parameters, request bodies, and response shapes.

Billing & Limits

The billing formula, the error catalog, and the operational limits.

Policies: Acceptable Use · Abuse & Takedown · Versioning & Deprecation

Overview

Reference

MCP

Policies

OpenAPI Reference

1. Get your API key

2. Provide your input video

3. Submit a job

4. Poll until terminal

5. Optional: completion webhooks

Next steps

Capabilities Reference

Billing & Limits

​1. Get your API key

​2. Provide your input video

​3. Submit a job

​4. Poll until terminal

​5. Optional: completion webhooks

​Next steps

Capabilities Reference

Billing & Limits

1. Get your API key

2. Provide your input video

3. Submit a job

4. Poll until terminal

5. Optional: completion webhooks

Next steps