Runbase

Runbase is a unified REST API for AI image and video generation. One key reaches models from OpenAI, Google, ByteDance, Kling, MiniMax and Qwen, with pay-as-you-go credits, automatic refunds on failures and published price comparisons with fal.ai and Replicate.

Runbase is a unified API for AI image and video generation. Instead of opening separate accounts with each model provider, developers use one key and one REST endpoint to reach models from OpenAI, Google, ByteDance, Kuaishou, MiniMax and Alibaba's Qwen. The catalog lists eight image-focused families such as GPT Image, Nano Banana, Seedream and Qwen Image, alongside video models including Seedance, Kling 3.0, Veo 3.1 Fast and Hailuo. In practice the workflow is short. A user signs up without a credit card, buys a $1 starter pack (one per account), creates an API key in the dashboard, and sends a POST request naming a model and its input, such as a prompt and resolution. The result is retrieved by polling. No SDK is needed, and each model has its own documentation page with code samples that can be copied into Claude or ChatGPT. A browser playground lets people try any model before writing integration code. Billing runs through a single credit wallet. Prices are shown per image, per clip or per second depending on the model, and the homepage publishes a comparison table against fal.ai and Replicate, claiming savings of up to 76% on some models while a few, like GPT Image 2.5, are only marginally cheaper. Failed generations are refunded automatically, so spend tracks successful outputs only. The audience is developers, startups and product teams building features that need generated visuals or short video, plus anyone who wants to compare models without juggling provider accounts. Runbase sits in the same space as fal.ai and Replicate, which it names directly as alternatives. Its pitch is price and consolidation rather than its own models, so it suits teams that value a lower per-generation cost and a single integration point more than a deep platform of training or hosting tools. Requests for unsupported models can be sent by email to the support address.