Lowest-cost media generation API

ModelRunner provides a unified API for generative AI models across image and video. Instead of managing multiple provider integrations, you access Google, ByteDance, and other providers through a single endpoint. This simplifies development, reduces integration overhead, and gives you flexibility to switch between providers or models without changing your code.

ModelRunner supports models across four categories: text-to-image, image-to-image, text-to-video, and image-to-video. We integrate with leading providers including Google (Veo 3.1, Imagen) and ByteDance (Seedance, Seedream), as well as open-source models like FLUX running on serverless compute. New models are added regularly, and you can test any model instantly in our Playground before integrating.

ModelRunner offers transparent, pay-as-you-go pricing with no hidden fees. Pricing varies by model type: per-second GPU time for serverless models, per-output for simple generation, or per-output-second for video (based on duration). All pricing is clearly displayed for each model, including GPU costs for serverless inference. You purchase credits via Stripe and only pay for what you use.

Yes. Every model on ModelRunner has an interactive Playground where you can test generation with full parameter control. You can explore example runs to see real inputs and outputs, then copy ready-to-use code snippets for your integration. The Playground mirrors the exact API behavior, so what you see is what you get in production.

Integration is straightforward: sign up, purchase credits, and create an API key. You can then make requests via our REST API or use our JavaScript SDK for a more streamlined experience. All models share a consistent interface—same authentication, same request/response patterns—so switching between models requires minimal code changes.

Yes. ModelRunner does not use your inputs or outputs for training. We support secure authentication including passkeys (WebAuthn), OAuth (GitHub, Google), and two-factor authentication, and your API keys are managed securely. Request inputs and outputs are stored so they appear in your dashboard history, and you control for how long: shorten retention on an individual request, set an account-wide default, opt out of storing payloads entirely with a single header, or delete a request's data — including the files it generated — whenever you want.

Yes. You can use ModelRunner-generated content for commercial purposes. However, licensing terms depend on the underlying model provider. We clearly document licensing for each model in our catalog. For open-source models, standard open-source licenses apply. For provider APIs like Google or ByteDance, their respective terms of service govern commercial use.

Yes. ModelRunner supports production applications requiring high throughput and reliability. Our async queue system handles requests at scale with real-time status updates via SSE. For enterprise needs like dedicated capacity, custom SLAs, or volume pricing, contact our sales team to discuss tailored solutions.

No. When you select a pro model, you always get that exact model—we never substitute it with a faster or cheaper alternative. Every request is processed by the model you chose, ensuring consistent quality and predictable results. This transparency is core to how ModelRunner operates: what you select is exactly what runs your generation.