RunningHub

RunningHub sells two things at once: a browser workspace for running and remixing ComfyUI graphs, and one API in front of hundreds of third-party models. The workspace pairs hosted graphs with studio tools built on them — rhTV's infinite canvas, AI Canvas, Quick Create. The API side covers hosted workflow calls, AI-app calls, workflow-JSON retrieval, resource and LoRA upload, and model endpoints with polling and webhooks. The vendor's own API page counts 322 video, 189 image, 47 audio and 16 3D models, naming Kling, Vidu, Seedance, Veo, Sora 2, Wan and Midjourney; you run no servers, and billing is on demand for actual usage, not reserved instances.

So what you buy is reach, not hardware. Prices are published per image, per second and per million token — the homepage advertises Seedance 2.5 from $0.036/sec, the API page shows nano-banana-2 at $0.027/image with a 60% membership discount — while deeper discounts sit behind tiers you must log in to read. Three things to watch: resold endpoints mean the licence and data terms behind each model are somebody else's; access is scoped by key type, and the docs say the LLM API only takes Enterprise-Shared keys, so a personal key reaches less; team accounts are a billing and quota layer over personal logins. It suits teams prototyping multi-model pipelines without ten vendor contracts, and the interface now offers at least ten languages, so calling it Chinese-only is stale. If you want a bare GPU you control, this is not that.