Popcraft API
Every leading AI model, behind one endpoint.
Ship image, video, audio, and avatar generation inside your own product. Popcraft handles the queueing, fairness, retries, and billing — you send one prompt and get a result.
Available to enterprise organizations · Charged only for successful generations
# one prompt in — a job id back, instantly
POST /v1/images/generations
{
"model": "nano-banana-2",
"prompt": "a tiny origami fox on a desk, soft morning light"
}
202 Accepted
{ "id": "job_9f2c1a", "status": "queued", "credits_reserved": 6 }Image
Text-to-image and reference-guided generation at production resolutions, from fast drafts to flagship quality.
Video
Cinematic text-to-video and image-to-video with explicit duration and resolution — priced per second, quoted up front.
Audio
Speech, sound effects, and music. Durations are explicit, so the credit quote is exact before a single frame is rendered.
Avatar
Drive a still portrait with audio or text-to-speech to produce a talking avatar video — one call, one clip.
Built for production
The hard parts are already solved.
Generation is slow, bursty, and expensive to get wrong. The platform is engineered so one tenant's flood never delays another's single job.
Leak detection, built in
Every API key is watched for volume and velocity spikes and access from unfamiliar countries. Anything suspicious alerts the owner instantly — so a leaked key is caught in minutes, not on the invoice.
Provider limits, absorbed
Per-kind concurrency caps and per-model rate budgets keep you inside provider limits automatically. Backpressure is a clean 429, never a silent drop.
Durable by default
Every job is persisted the instant it's accepted. Deploys, restarts, and instance failures never lose a job or bill you twice.
Async from the first byte
Submit returns in milliseconds with a job id. Poll for state, or let a webhook push the result the moment it's ready.
Signed webhooks
Terminal events are delivered with an HMAC signature and retried on failure — so you can skip polling entirely and still never miss a result.
A live model catalog
Query available models, their capabilities, limits, and credit price at runtime. New models appear without a code change on your side.
Plans
Start on defaults. Scale on Enterprise.
No maze of tiers. Every organization starts on the same sensible defaults and pays only for what it generates. When you need more throughput, Enterprise turns each limit into a number we set with you — concurrency first.
Every organization · day one
Default
Everything you need to build, launch, and self-serve.
Pay as you go
credits only · no commitment
- Concurrency10 image · 4 video · 6 audio · 2 avatar in-flight
- Request rate60 / min
- Queue depth50 waiting per kind
- DeliveryPoll or signed webhook
- Spend controlPer-key limits & org ceiling
When you scale · negotiated
Enterprise
The same platform, with every limit turned up to your number.
Let's talk
negotiated rate & limits
- ConcurrencyRaised to your throughput — name the number
- Rate & capacityHigher RPM · dedicated per-model budgets
- CommercialInvoicing · net terms · SLA
- AssuranceSecurity review · DPA · named engineer
- Spend controlCustom ceilings, still owner-controlled
Room to negotiate
Concurrency is yours to raise.
The Default numbers are a floor, not a ceiling. On Enterprise, every one of these — how many jobs run at once, how fast you can call, how deep the queue goes — is set with you and applied per organization. No code change and no new keys on your side.
| Dimension | Default | On Enterprise |
|---|---|---|
| Image jobs in-flight | 10 concurrent | Contact us to raise |
| Video jobs in-flight | 4 concurrent | Contact us to raise |
| Audio jobs in-flight | 6 concurrent | Contact us to raise |
| Avatar jobs in-flight | 2 concurrent | Contact us to raise |
| Requests / min | 60 | Contact us to raise |
| Queue depth | 50 waiting per kind | Contact us to raise |
| Per-model capacity | Shared pool | Contact us for dedicated |
| Monthly credit ceiling | You set it | Contact us to raise |
These are real controls, applied per organization the moment your plan changes. Raising throughput is a conversation, not a migration.
Spend & safety
The owner holds every lever.
Handing a developer a key doesn't mean handing over the budget. The owner caps spend, watches every credit, and shuts down anything that looks wrong — without opening a ticket.
Cap the spend
Set a monthly ceiling for the whole organization and a credit limit on each key. Both are enforced at submit — an over-limit request is refused before it costs a single credit.
Catch abuse early
Every key is watched for volume and velocity spikes and calls from unfamiliar countries. The owner is alerted by email and in-console the moment a key looks leaked — revoke it in one click.
No surprise invoices
Failed, cancelled, and timed-out jobs are never billed. With caps and alerts on top, a leaked key simply cannot run up a bill you did not sign off on.
Billing
You pay for results, not attempts.
Credits are reserved when you submit and deducted only when a job completes. Failed, cancelled, and timed-out jobs cost nothing — so a refund is something you'll simply never need.
Exact quotes up front: every request is priced from the live catalog before it runs, and expected-leftover math on every status read means you always know what you can still afford.
The questions procurement asks.
Usage is metered in credits and charged against your organization's monthly cycle — one shared balance, not a separate wallet per developer. Every request is priced up front from the live catalog, and you're charged only when a job completes. Failed, cancelled, and timed-out jobs cost nothing.
Yes, at two levels: an organization ceiling per cycle and a per-key credit limit, so a single integration can't run away with the budget. Both are enforced at submit — an over-limit request is refused before it costs a credit. The owner controls both.
The console shows live usage by API key and by team member, with a recent jobs-and-charges ledger. Export a CSV statement by job or by model for any date range. Enterprise adds monthly invoicing with net terms.
Each model kind has an in-flight cap plus per-key caps and a bounded queue. A fair scheduler round-robins across keys, so one customer's burst of 200 jobs never starves another's single job. Extra work queues rather than failing — and on Enterprise the caps are raised to your throughput.
Requests are governed by per-key and per-organization rate buckets. The Default numbers are a floor; concurrency, request rate, queue depth, and per-model capacity are all per-organization overrides we apply instantly — no redeploy, no new keys.
You get a clean 429 with backpressure — never a silent drop. Jobs already accepted keep flowing through the queue; you pace new submissions or ask us to raise the ceiling.
Async. Submit returns in milliseconds with a job id. Then either poll the generation endpoint or let a signed webhook push the result the moment it's ready.
Terminal events carry an HMAC signature you verify against your per-organization secret, and are retried on failure. So you can skip polling entirely and still never miss a result.
Built-in abuse detection watches every key for volume and velocity spikes and activity from unfamiliar countries, and alerts your team by email and in-console the moment something looks off. Revoke in one click — and spend caps mean the blast radius is bounded even before you react.
Yes. Grant a member console access with self-scope — they manage only their own keys and see only their own usage, never org billing or other members' keys. Owners and admins see everything.
The API is enabled per organization. Your account manager switches it on and sets your limits; from there owners self-serve keys, docs, usage, and billing in the console. Most teams are calling production models the same day.
Getting started
Access is granted per organization.
The API is available to enterprise organizations. Once your organization is enabled, owners and admins self-serve everything from the dashboard.
01
Get enabled
Your account manager enables API access for your organization and sets up your dedicated credit pool.
02
Mint keys, read the docs
Create scoped API keys for your developers from the console — and read the full interactive reference with copy-paste examples.
03
Ship it
Submit, poll, and render. Watch usage and billing update live in the console as your product generates.