API

1. Create a key

Create a key in API Key and save it as a server-side environment variable.

export FLUX_API_KEY="YOUR_FLUX_API_KEY"

2. Check your connection

This request reads your balance and does not spend credits.

curl https://api.flux-context.org/api/v1/balance \
  -H "Authorization: Bearer $FLUX_API_KEY"

3. Create a generation

Choose a model to see its parameters and a complete request example.

4. Poll and download

Save the returned request ID and query approximately every 10 seconds. Background work continues even when you stop polling.

curl https://api.flux-context.org/api/v1/generations/REQUEST_ID \
  -H "Authorization: Bearer $FLUX_API_KEY"

When status is succeeded, download each URL in results. Response excerpt:

{
  "id": "REQUEST_ID",
  "status": "succeeded",
  "results": [
    {
      "url": "https://api.flux-context.org/api/v1/files/SIGNED_TOKEN",
      "url_expires_at": "2026-09-10T11:42:18Z",
      "expires_at": "2026-10-10T10:42:18Z",
      "expired": false,
      "media_type": "image",
      "content_type": "image/png",
      "size_bytes": 1234567,
      "width": 1024,
      "height": 1024
    }
  ]
}

Stop polling when the status is succeeded or failed. For failed tasks, read error and billing for the reason and final credit settlement.

Download links last one hour. Query again for a fresh link. Files are retained for 30 days after transfer; file expiry does not change generation or billing records. Video results use media_type video and content_type video/mp4; save them with a .mp4 extension. Downloads support HEAD and single byte-range requests for playback and resuming.

curl --fail --location "$RESULT_URL" --output result.png

Credits and settlement

No subscription is required. API generations spend subscription credits first, then purchased credits. Gift credits cannot be used, including for additional video charges. GET /balance returns credits_available for these eligible pools only; breakdown.giftRemaining is informational and is not included. Refunds follow the original charge.

Fixed requests are charged upfront. Seedance 2.5 editing reserves credits and settles automatically after generation. Billing includes credits_precharged, credits_additional, credits_refunded, credits_final and settlement_status. credits_charged remains the total debited before refunds. credits_final is null while pending. If settlement_mode is precharge_retained, the precharge remains debited and results are available while billing awaits review.

Missing final billing data stays pending and is retried in the background. No manual log checks are required. Results are available after settlement. File expiry does not refund a successful generation.

Reference media

Supply public HTTPS URLs, without duration fields. Flux measures MP4/MOV video and MP3/WAV/M4A audio before charging. References must stay available until generation finishes. Requests use up to 3 concurrent probes, a 20-second deadline, an 8 MiB metadata budget per file and 32 MiB in total. Range support is recommended; small files also work without it.

INVALID_REFERENCE_MEDIA (HTTP 400) means media could not be read reliably or exceeded a model constraint. No task was created or credits charged. Check the URL, file format and duration, then retry.

Common request headers

A network timeout does not prove creation failed. Use the optional retry header when automatically retrying POST requests. The Python example handles it for you.

Errors & limits

Each HTTP attempt has a request_id and X-Request-Id header. The generation id remains stable across retries using the same Idempotency-Key and unchanged body. Use the Flux generation id to query status; provider_task_id is an optional support reference, not a Flux lookup key.

Errors keep the Flux error.code and add source and provider details when available. Provider business codes, failure codes and HTTP status are distinct from the Flux HTTP status. A successful status lookup does not mean the generation succeeded.

Read the error explanation and billing state before retrying. Correct invalid inputs; wait for rate limits; query the original task when the outcome is unknown. Upstream account or configuration errors require Flux support, not a new Flux API key or additional Flux credits.

For support, copy troubleshooting information from API Logs. Include the HTTP request ID, Flux generation ID and provider task ID when available. Never share API keys, Authorization headers or private input data.

These are default limits. Your effective limits appear in Overview and GET /balance. All keys share the account limit. HTTP 429 includes Retry-After: 60 and a limit object with scope and value. For TOO_MANY_TASKS, wait for existing generations to finish; 60 seconds does not guarantee a slot.

SUBMISSION_NOT_CONFIRMED is not a confirmed failure. Keep the original request ID and query its status. Temporary query errors do not trigger refunds.