Runware

以下为收录时的原文，语言与页面不同，未做翻译。

Runware offers an API and web Playground for image, video, and audio generative inference—supporting text-to-image, image-to-image, inpainting, outpainting, ControlNet, custom model uploads, background removal, upscaling, automatic captioning, and low‑latency batch execution.

官网介绍

Build with generative AI on a unified inference API. Image generation, video generation, audio, 3D, and large language models. 400K+ models, managed infrastructure, usage-based pricing.

官网要点

• Up to 10× lower cost per generation
• Edit, upscale and background removal built in
• API Gateway
• Image Generation
• Video Generation
• Audio Generation

官方常见问题

What is the Sonic Inference Engine®?

A fully custom hardware and software stack, built specifically for AI inference. We tune it from BIOS and kernel up, run models on hardware we own, and keep them preloaded across regions. Because we operate the engine end to end, you get higher throughput and lower latency than generic cloud GPUs, at lower cost.

What does “one API for all AI” mean in practice?

One auth, one endpoint, one bill. Switching model is a string change. Image, video, audio, 3D, vision and LLM chat all hit the same POST /v1. Connect over REST, or use WebSockets for persistent connections. If you have your own weights, you can upload LoRAs, checkpoints, safetensors and LyCORIS through Model Upload.

How does Runware's pricing work?

Pay per request. No subscriptions, no commitments. Open-source models are billed on optimised compute time, so faster generations automatically cost less. Closed-source and partner models are fixed per-request, at rates we negotiate down with the provider. In practice, open-source typically runs up to 10× cheaper and 40% faster than other providers; closed-source lands 10–40% below the published r

What models does Runware support?

400K+ open and proprietary models, including Flux 2, Veo 3.1, Kling, Seedance, ElevenLabs, Claude Opus 4.7, GPT-5.4, Llama, Qwen and DeepSeek, plus a wide range of vision models. You can also bring your own LoRAs, checkpoints, safetensors and LyCORIS through Model Upload. Nothing is artificially restricted for speed, and there's no hidden caching that alters outputs.

Can I use Runware for commercial projects?

Official models on Runware include commercial usage rights under our partner agreements, so you can use leading models in production without worrying about separate licence fees. For community models, commercial use is governed by the licence published by the model creator. We always link to the source so the terms are easy to review.

Is my data private and secure?

Inputs and outputs are never used for training. Encrypted in transit. SOC 2 and ISO 27001 certified, GDPR aligned. Generated content is automatically purged from our servers unless you explicitly opt in to storage. Your data always belongs to you and is never reused, resold, or used for any other purpose.