POST/v1/generate/video
Video generation
Submit text or image prompts, choose any of 7+ video models, set duration and aspect ratio. Returns a job ID for async polling or webhook delivery.
The Kaiber API gives developers programmatic access to every model in the Kaiber stack — Veo 3.1, Kling 3.0, Luma Ray, Runway Gen-4.5, Flux Pro, plus Cuts auto-edit and custom-trained models. One integration, fifteen-plus models, no per-vendor contracts.
The Kaiber API exposes the same model stack and creative tooling that powers Superstudio wrapped in a clean REST interface designed for developers building video AI into their own products.
Most generative video APIs are single-model offerings. You sign up for Runway's API and you get Runway models. You sign up for Pika and you get Pika. You sign up for Luma and you get Luma. Each contract is separate, each billing system is separate, each authentication scheme is separate. By the time you've integrated three or four model vendors to give your product enough creative range, you're managing a small constellation of API contracts and your finance team is asking why there are so many AI invoices.
The Kaiber API takes a different approach. One integration, every model. Through a single REST endpoint, you can generate video on Veo 3.1 (Google's native-audio model), Kling 3.0, Luma Ray, Runway Gen-4.5, MiniMax Hailuo, Mochi, plus image generation on Flux Pro, Recraft, and Stable Diffusion. You can run a custom-trained LoRA model. You can trigger a Cuts auto-edit. You can upscale to 4K. All of it, one API key, one billing system, one set of credits.
The API is positioned for builders companies embedding generative video into their own products, agencies running custom workflow tooling for clients, and large content operations automating production pipelines. It's not a self-serve developer tier; access is provisioned through the Visionary plan (Kaiber's enterprise tier) and includes dedicated technical onboarding. We'll cover access in section 07.
The Kaiber API is currently in private access for Visionary-tier customers. Self-serve developer access is on the public roadmap. To get early access or evaluate the API for your product, contact support@kaiber.ai with a brief description of your use case and expected volume.
Under the hood, the API is a thin orchestration layer over Kaiber's existing model integrations. When you submit a video generation job to /v1/generate/video with model: "veo-3.1", your request gets routed to Kaiber's Veo integration, runs on Google's infrastructure, and returns a finished video URL alongside metadata. When you submit the same prompt with model: "kling-3.0", the same request flows to Kuaishou's Kling backend instead. From your application's perspective, switching between premium video models is a one-string change no separate auth, no separate contract, no different request shape.
Every Kaiber capability that's available in the web platform is exposed through a corresponding API endpoint. Here's the full surface area, organized by what you'll build with.
Submit text or image prompts, choose any of 7+ video models, set duration and aspect ratio. Returns a job ID for async polling or webhook delivery.
Generate stills via Flux Pro, Recraft v3, or Stable Diffusion 3.5. Returns image URLs, supports batch requests of up to 8 images per call.
Re-skin existing footage in any AI style. Upload a video, choose a style preset or custom model, get back the transformed clip.
Submit audio plus visuals, get back a beat-synced video. Programmatic access to the same Cuts engine that powers the Kaiber web app — including auto-clip and lyric overlays.
Trigger a LoRA training run from a set of reference images. Trained models become available across all video and image generation endpoints.
Push images and videos to higher resolution via Magnific (image) and Topaz (video) integrations. 1080p, 2K, or 4K targets supported.
Check the status of any submitted job. Returns queued, processing, completed, or failed status plus output URLs and metadata when ready.
List every available model with current pricing, supported parameters, and availability status. Useful for building dynamic UIs in your own app.
Register webhook URLs that receive POST notifications when jobs complete. Eliminates polling for long-running generation tasks.
Authentication uses standard Bearer tokens. Keys are scoped, rotatable, and rate-limited per environment. Here's the full flow.
Email support@kaiber.ai with your use case and expected volume. Visionary-plan customers can request keys directly from their dashboard.
From your dashboard, generate keys scoped to specific endpoints (video, image, cuts, training). Separate keys for production, staging, and CI.
Pass the key as Authorization: Bearer $KAIBER_API_KEY on every request. HTTPS required; HTTP requests are rejected.
Rotate keys at any time without downtime - old and new keys overlap for 24 hours. Compromised keys can be revoked instantly.
Standard API access on Visionary plan: 60 requests per minute on read endpoints, 10 concurrent generation jobs. Custom limits available on enterprise contracts. Rate-limit headers (X-RateLimit-Limit, X-RateLimit-Remaining, X-RateLimit-Reset) return on every response.
Generate a 5-second video on Veo 3.1 in any of three languages. Authentication, request shape, and error handling are identical across SDKs - so switching between languages is mechanical.
Submit a text prompt to Veo 3.1 and get a job ID back. The actual video URL becomes available once the job completes typically in 20-60 seconds.
Generation jobs run async. Either poll the job endpoint until status is completed, or register a webhook to receive a callback when the job finishes (recommended for production).
Submit an audio file plus a set of visuals, get back a beat-synced video — the same Cuts engine that powers the web app, exposed via API.
The full list of models exposed via the Kaiber API, with current credit costs. API costs match the in-app costs you're not paying a developer-tier markup.
| Model ID | Provider | Type | Credit Cost |
|---|---|---|---|
| veo-3.1 | Video + native audio | 36 cr/sec | |
| kling-3.0 | Kuaishou | Video, long takes | ~10–15 cr/sec |
| luma-ray-3 | Luma Labs | Cinematic video, HDR | ~12 cr/sec |
| runway-gen-4.5 | Runway | Filmmaker-grade video | ~14 cr/sec |
| hailuo-01 | MiniMax | Fast iteration video | ~8 cr/sec |
| mochi-1 | Genmo | Stylized video | ~10 cr/sec |
| flux-pro | Black Forest Labs | Photorealistic image | ~3–5 cr/image |
| recraft-v3 | Recraft | Design / typography | ~3 cr/image |
| sd-3.5 | Stability AI | Versatile image | ~2 cr/image |
| topaz-vid-upscale | Topaz Labs | Video 4K upscale | ~40 cr/clip |
| magnific-img-upscale | Magnific | Image hi-res upscale | ~10 cr/image |
| cuts-render | Kaiber | Beat-sync auto-edit | 0 cr (free) |
| custom/<your_model_id> | Kaiber | Trained LoRA model | 500 cr (one-time) |
The credit costs above are identical to the in-app costs in Kaiber Canvas. There's no API surcharge or developer-tier multiplier. The same 5-second Veo 3.1 generation that costs 180 credits when you click "Generate" in Superstudio costs 180 credits when you submit it via POST /v1/generate/video. Beat Sync (cuts-render) is free in both contexts.
Four representative use cases across content platforms, agencies, music labels, and creative tooling — drawn from real Visionary-tier customers.
Social and creator platforms embed Kaiber-powered video generation directly into their app — users type a description and get back a styled video without ever leaving the platform. Kaiber handles model selection, queueing, and delivery; the platform handles UI and user accounts.
A creative agency builds an internal tool that takes a client's brief, generates 5–10 video variations across multiple AI models in parallel, and assembles them into a pitch deck. What used to be a 3-day storyboard process becomes a 30-minute review meeting.
A label automates the production of Spotify Canvas loops for every track on every release. The pipeline ingests track audio + a brand visual reference, generates 16:9 and 9:16 visuals on the appropriate model, and queues them for QA before shipping to Spotify.
A video-editing software vendor adds a "Generate B-roll with AI" panel to their NLE. Editors describe what they need ("aerial drone shot of a coastline at sunset"), the plugin calls the Kaiber API in the background, and finished clips drop straight into the project bin.
API access is currently provisioned through the Visionary plan with custom pricing tied to your expected volume. Here's the path from intro email to first generation.
The Kaiber API is included on the Visionary plan along with white-glove technical onboarding, a dedicated account manager, custom credit allocation matched to your volume, SSO for team access, and SLA guarantees. Pricing is custom-quoted based on expected monthly credit consumption, number of seats, and integration support requirements.
Typical Visionary engagements start in the low four figures per month and scale from there. Companies running production at meaningful volume (tens of thousands of generations per month) can negotiate volume discounts off standard credit rates. Smaller teams running lower-volume integrations may be a better fit for credit packs purchased through the Pro plan.
support@kaiber.ai with use case & volume
support@kaiber.ai.veo-3.1 at 36 credits/second). Overage beyond your allocation is billed at standard credit-pack rates. Detailed usage reports are available in the dashboard, broken down by model, endpoint, and time window.400 Bad Request with a policy-violation error code.POST /v1/training/start endpoint accepts a set of reference images and triggers a LoRA training run. Training takes 10–30 minutes and costs ~500 credits per run. Once complete, the trained model is referenced by custom/<your_model_id> in subsequent generation requests and is private to your organization.POST /v1/webhooks or per-job in the generation request. When a job completes (or fails), Kaiber sends a signed POST to your URL with the job result payload. Signatures use HMAC-SHA256 with a per-org webhook secret — verify them before acting on the payload to prevent spoofed callbacks.@kaiber/sdk on npm), Python (kaiber on PyPI), and Go (github.com/kaiber-ai/go-sdk). All three SDKs cover the full API surface and handle authentication, polling, and webhook signature verification automatically. The raw REST API can be called from any language that can make HTTPS requests.veo-3.1, kling-3.0) and provides at least 90 days notice before deprecating any specific version. New model versions ship with new IDs (e.g. veo-3.2) so existing integrations don't break automatically. The GET /v1/models endpoint always reflects current availability and any deprecation warnings.https://sandbox-api.kaiber.ai) and reduced-cost test credits. Sandbox responses are real generations on real models — there's no mock mode — but they're tagged in your usage reports and can be billed against a separate test allocation.The Kaiber API is in private access for Visionary-tier customers. Get in touch with use case, expected volume, and integration timeline — typical responses within 1 business day.