Question 1

What is BabySea?

Accepted Answer

BabySea is AI inference infrastructure for generative media. It executes, routes, retries, bills, and delivers every image and video generation request through one consistent API and schema, across multiple inference providers, without requiring provider-specific orchestration in your own code. You integrate once. See the model schema or learn more why we built BabySea .

Question 2

How do I get started?

Accepted Answer

Sign up for free with no card required, every new account starts with $1 in welcome credits. Create an API key, verify it with `GET /v1/status` , then send a generation request to `POST /v1/generate/image/{model_identifier}` or `POST /v1/generate/video/{model_identifier}` . After initialization, retrieve the finalized record from `GET /v1/content/{generation_id}` . Start with the quickstart guide or test it live in the playground .

Question 3

How long does generation take?

Accepted Answer

Image jobs usually finish in seconds, while video jobs take longer depending on the model, provider, duration, and output settings. BabySea can fail over across providers when more than one backend is configured, so your application does not need custom retry and recovery logic. See model pricing for model-specific cost and capability details, and error reference for failure behavior.

Question 4

What is the unified schema?

Accepted Answer

BabySea exposes one request and response model across providers. Instead of adapting to provider-specific inputs like raw width and height fields, provider enums, or different file conventions, you send one normalized shape using fields like `generation_ratio` , `generation_input_file` , and `generation_provider_order` . BabySea translates that request, executes it, and returns one consistent lifecycle record. See the model schema .

Question 5

What providers does BabySea support?

Accepted Answer

BabySea currently orchestrates inference across Replicate, Fal, BytePlus, Cloudflare, Black Forest Labs, OpenAI, and more. Models from additional labs can still appear in the catalog when BabySea routes them through those infrastructure providers. See the model schema for current provider coverage and model mapping.

Question 6

What image models are available?

Accepted Answer

The image catalog covers fast drafts, high-resolution outputs, text rendering, photorealism, editing, and more. New models are added as BabySea expands provider coverage and normalizes them into the shared schema. See model pricing for the current catalog and pricing.

Question 7

What video models are available?

Accepted Answer

The video catalog spans fast clips, cinematic output, and models that support image-to-video and audio-enabled workflows. Video usage is priced by output duration and can also vary by resolution and model family. See model pricing for the live catalog and pricing.

Question 8

How do I choose the right model?

Accepted Answer

Choose based on output type, latency, provider coverage, supported fields, and cost. Compare models on the model pricing , then call `GET /v1/estimate/{model_identifier}` before generating so you can preview exact credit usage. The playground is the fastest place to test models side by side.

Question 9

What do I get for free?

Accepted Answer

Every new user receives $1 in welcome credits automatically. The Free plan also includes 10 generations per minute, 30 API requests per minute, signed webhook delivery with retries, IP access control, the metrics dashboard, and 1 hour of file retention. See the credits overview and billing plans for current plan details.

Question 10

How does pricing work?

Accepted Answer

BabySea uses a subscription-plus-credits model. Your plan controls limits, retention, and feature access, while usage is billed separately through credits based on model, output type, and generation settings. Paid plans are Starter ($9/month), Pro ($29/month), Scale ($99/month), and Enterprise ($199/month), with yearly pricing set at 10x the monthly rate. See pricing and billing plans for the full breakdown.

Question 11

What credit packs are available?

Accepted Answer

Six one-time credit packs are available for accounts with an active subscription: $10, $25, $50, $100, $500, and $1,000. Only the Primary Owner can purchase them, and purchased credits accumulate on the shared account balance. See the credit packs guide and pricing .

Question 12

Do credits expire?

Accepted Answer

No. Credits remain on your balance until they are used. Every reserve, charge, refund, and credit addition is written to the credit ledger so the balance is auditable at all times. See the credits overview and credit transactions .

Question 13

How much does each generation cost?

Accepted Answer

Cost depends on the model, output type, and settings. Image models are typically priced per generation, while video models are priced per second of output. Use `GET /v1/estimate/{model_identifier}` to preview exact credit usage before you run a job. See model pricing and credit usage .

Question 14

How do I integrate with the API?

Accepted Answer

You can integrate through the TypeScript SDK with `npm install babysea` or call the REST API directly. Either way, BabySea handles provider routing, retries, billing, and delivery behind one API surface, so you do not need provider-specific SDKs or orchestration code. Start with the docs and SDK overview .

Question 15

What can I do in the Playground?

Accepted Answer

The Playground is a browser-based API console for generation and generation management. You can choose image or video models, keep your API key in memory, send live `/v1/generate` requests, inspect the JSON response, then look up, cancel, or delete a generation by `generation_id` . See the playground guide and content management .

Question 16

How do I verify my API key and monitor usage programmatically?

Accepted Answer

Use `GET /v1/status` to smoke-test an API key and inspect metadata like key prefix, creation time, and expiry. Use `GET /v1/usage` for aggregated request, generation, endpoint, and provider summaries over a lookback window. Account and billing summaries are also available through `GET /v1/user/account` and `GET /v1/user/billing` . See the SDK methods and billing overview .

Question 17

How do API keys work?

Accepted Answer

BabySea API keys use the `bye_` prefix and authenticate every public `/v1` request with a Bearer token. The full key is shown only once at creation, only the key prefix and a bcrypt hash are stored. At creation, you choose a scope preset and can optionally attach an IP allowlist. Rotation creates a replacement key with inherited permissions and a grace period before the old key expires. See the API keys and key rotation guide .

Question 18

What API key scopes are available?

Accepted Answer

Each API key can be created with one of four presets: Full Access, Generate Only, Read Only, or Monitor Only. These map to enforced route permissions across generation, account, health, usage, and library endpoints. Scopes are immutable after creation, so changing permissions means issuing a new key. See the API scopes for exact scope behavior.

Question 19

How do webhooks work?

Accepted Answer

BabySea emits signed lifecycle events for `generation.started` , `generation.completed` , `generation.failed` , and `generation.canceled` . Deliveries are retried automatically with exponential backoff, and failed destinations can queue events in the dead letter queue instead of losing them. See the webhook overview and retry & delivery .

Question 20

Should I poll for results or use webhooks?

Accepted Answer

Both are supported. Polling is simplest when you are prototyping: call `GET /v1/content/{generation_id}` until the generation finishes. In production, most server-side integrations prefer webhooks so BabySea can push the signed lifecycle event as soon as the job state changes. See the quickstart guide and webhook overview .

Question 21

How do I verify webhook signatures and operate webhook endpoints?

Accepted Answer

Every webhook request includes `X-BabySea-Signature` , `X-BabySea-Event` , `X-BabySea-Delivery-Id` , and `X-BabySea-Timestamp` . You can verify the HMAC manually or use the SDK helper, which also checks the default timestamp tolerance window. In the dashboard, owner-level users can replay deliveries, rotate secrets, disable endpoints, and drain queued events after re-enabling a webhook. See the signature verification and retry & delivery .

Question 22

What regions are available?

Accepted Answer

BabySea provides three default API base URLs: US at `https://api.us.babysea.ai` , EU at `https://api.eu.babysea.ai` , and APAC at `https://api.jp.babysea.ai` . The SDK supports region selection, and Enterprise accounts can attach one additional custom API domain that works alongside the default endpoint. See the SDK overview and domains .

Question 23

How do custom API domains work?

Accepted Answer

Custom API domains are an Enterprise feature. Your default BabySea endpoint always remains available, and Enterprise accounts can attach one additional domain such as `api.yourcompany.com` . Owner-level users add the domain in the dashboard, configure the required DNS record, verify SSL, and then use the custom `/v1/` base URL alongside the standard endpoint. See the custom domain setup and domain lifecycle .

Question 24

How does failover routing work?

Accepted Answer

BabySea executes every request through a deterministic provider chain. If one provider times out or returns a mapped upstream failure, BabySea reroutes the request to the next provider without forcing you to manage recovery logic in your own code. See the model schema and error reference .

Question 25

How do provider timeouts work?

Accepted Answer

Each provider attempt runs inside an enforced timeout window. Image routes use shorter provider timeouts, while video routes allow much longer execution windows because video models can take significantly more time. If an attempt exceeds the timeout or becomes unhealthy, BabySea moves to the next provider in the configured chain. See the error reference and API and inference terms .

Question 26

What happens if a provider fails?

Accepted Answer

BabySea automatically fails over to the next provider in the configured chain when the current provider times out or returns a mapped upstream error. If every configured provider fails, the API returns a retryable provider error and failed generations are refunded back to your balance. See the error reference and credits overview .

Question 27

What happens if generation fails?

Accepted Answer

Reserved credits are refunded automatically for failed generations. Pending or processing generations can also be canceled, but very fast jobs may finish before the cancel request is applied, in which case the API returns a conflict and the completed job remains charged. Structured BSE errors include category, machine-readable code, and retryability. See the error reference and content management .

Question 28

Can I cancel or delete generations after I submit them?

Accepted Answer

Yes. Pending or processing jobs can be canceled with `POST /v1/content/generation/cancel/{generation_id}` , and stored generation records plus output files can be removed with `DELETE /v1/content/{generation_id}` . These actions are also available in the Playground management flow. See the content management and SDK methods .

Question 29

How is my data secured?

Accepted Answer

Current controls include scoped API keys, per-key IP allowlists, HMAC-signed webhooks, bcrypt-hashed API keys, row-level security in the application data layer, audit logging, and security headers on public routes. Operational permissions are also enforced by role in the dashboard so sensitive actions are not exposed to every member. See the trust center and privacy policy .

Question 30

What are the rate limits?

Accepted Answer

Rate limits are enforced per account in a shared 60-second sliding window across all API keys on that account. The current limits are Free: 10 generations/min and 30 API requests/min, Starter: 20 and 60, Pro: 50 and 150, Scale: 100 and 300, Enterprise: 200 and 600. When you hit a limit, the API returns HTTP 429 with a `Retry-After` header that the SDK can respect automatically. See pricing and API and inference terms .

Question 31

What analytics are available?

Accepted Answer

The dashboard covers plan and credit usage, generation outcomes, model and provider usage, request and rejection counts, webhook delivery performance, latency percentiles, and error breakdowns. Historical charts survive normal cleanup because aggregate analytics snapshots are preserved separately from short-lived raw files. See activity retention and export and log retention .

Question 32

Can I export operational logs and credit history?

Accepted Answer

Yes. CSV and JSON exports are available for operational tables like activity and logs, and the credit ledger keeps a full transaction trail across reserves, charges, refunds, and credit additions. That makes audits, incident review, and reconciliation much easier. See the log export and credit transactions .

Question 33

Do you support team accounts?

Accepted Answer

Yes. Team workspaces use three roles: Primary Owner, Owner, and Member. Primary Owners control billing and ownership transfer, Owners manage most operational features, and Members are largely view-only. Credits and operational history are shared at the account level. See the member roles and billing plans .

Question 34

Can I purchase more credits anytime?

Accepted Answer

Yes. Accounts with an active subscription can purchase one-time credit packs from $10 to $1,000 directly from Billing, and only the Primary Owner can complete the purchase flow. See the credit packs guide and pricing .

Question 35

How long are generated files stored?

Accepted Answer

Output file retention depends on plan: Free (1 hour), Starter (12 hours), Pro (1 day), Scale (5 days), and Enterprise (7 days). After the retention window ends, files are automatically deleted from storage while the generation metadata remains in your logs for tracking and analytics. See pricing and data lifecycle terms .

Question 36

How long are generation logs kept?

Accepted Answer

Generation logs keep the record of each request, including prompt, model, provider, status, timing, and cost. Retention depends on plan: Free (1 day), Starter (7 days), Pro (30 days), Scale (180 days), and Enterprise (1 year). Aggregate analytics snapshots are preserved separately so trend charts can outlive raw log cleanup. See pricing and data lifecycle terms .

Question 37

How long are API request logs kept?

Accepted Answer

Every API request is logged with endpoint, method, status, and operational metadata. Retention follows the same plan-based windows as most account logs: Free (1 day), Starter (7 days), Pro (30 days), Scale (180 days), and Enterprise (1 year). See pricing and log retention .

Question 38

How long are webhook event logs kept?

Accepted Answer

Webhook delivery logs track event type, destination, delivery attempts, and response status. They follow the same plan-based retention windows: Free (1 day), Starter (7 days), Pro (30 days), Scale (180 days), and Enterprise (1 year). See pricing and data lifecycle terms .

Question 39

How long are activity logs kept?

Accepted Answer

Activity logs capture account-level changes like API key events, webhook updates, billing changes, and allowlist edits. Retention follows the same plan windows as operational logs: Free (1 day), Starter (7 days), Pro (30 days), Scale (180 days), and Enterprise (1 year). See pricing and activity retention and export .

Question 40

How long is the credit audit trail kept?

Accepted Answer

The credit audit trail records deposits, purchases, reserves, charges, and refunds. It is retained longer than most operational logs: Free (30 days), Starter (90 days), Pro (180 days), Scale (1 year), and Enterprise (1 year). See pricing and data lifecycle terms .

Question 41

Can I export or delete my data?

Accepted Answer

Yes. CSV and JSON exports are available for table-based account areas like Activity, and generated content can be removed through `DELETE /v1/content/{generation_id}` or through the Playground management tools. Personal account deletion is also available through an OTP-verified destructive flow. See account access and deletion and contact us .

Question 42

How does BabySea handle data privacy?

Accepted Answer

BabySea records consent decisions server-side, exposes privacy and cookie policy pages, and supports account deletion flows that scrub PII from activity logs before removal. Analytics and marketing preferences are handled as explicit opt-in categories rather than one blanket toggle. See privacy policy and account access and deletion .

Question 43

Where can I find legal, policy, and service information?

Accepted Answer

Product documentation lives on the docs , while legal and policy documents live on the terms site . That includes terms of use, privacy policy, service terms, cookies policy, and AI-related policy pages.

Question 44

Where can I review BabySea's security and compliance posture?

Accepted Answer

The trust center provides the high-level view of security controls, infrastructure, privacy posture, and legal documentation. For more detail on how account and request data are handled, see the privacy policy .

Question 45

What payment methods are accepted?

Accepted Answer

Billing runs through a hosted checkout and billing portal. The payment methods available to you are the ones surfaced there for your region and account, and subscriptions can be purchased monthly or yearly. See pricing and billing plans .

Question 46

Can I cancel or change my plan?

Accepted Answer

Yes. The Primary Owner can use the billing portal to update payment details, switch plans, download invoices, or cancel a subscription. Owners and Members can view billing information but cannot complete billing changes. See the billing overview and contact us .

Question 47

Can I get a refund?

Accepted Answer

Failed generations are refunded automatically through the credit ledger. For subscription disputes or refund requests outside of automatic generation refunds, see billing and credit terms and contact us .

Question 48

Does BabySea filter or moderate content?

Accepted Answer

Content moderation is primarily enforced by upstream model providers. BabySea preserves that provider behavior and returns clear structured errors when content is rejected, rather than introducing an opaque extra filtering layer of its own. See error reference and AI principles .

Question 49

What output formats are supported?

Accepted Answer

Image models typically return PNG or JPEG, depending on the model and requested parameters, while video models return MP4. The exact output options depend on the model schema, and generated files are returned as URLs during the retention window. See model pricing and model schema .

Question 50

What's on the roadmap?

Accepted Answer

New models continue to arrive as provider support expands, but the main focus remains infrastructure quality: faster failover, lower latency, stronger observability, and more reliable execution under production load. Check model pricing to see the latest catalog.

FAQ