Published Updated
In one or two sentences, before anything else.
RunningHub is a cloud platform for running ComfyUI workflows and AI apps on rented GPUs, made by the Chinese company Haima Cloud (海马云) and offered at runninghub.cn in China and runninghub.ai abroad. By 2026 it had grown into a creation suite with shared workflows, hosted video models, an agent canvas and APIs.
What it can do, one sentence each, and whether we actually tested it.
Build, edit and run ComfyUI workflows in the browser on RunningHub's machines, which its plan page describes as 90-series GPUs with 24 GB or 48 GB of memory.
RunningHub says it preinstalls 13,681 nodes and more than 100,000 models (its own figures, May 2026).
Workflow authors publish a workflow, or wrap it as an AI app with a few simple inputs, for others to run on their own coins.
RunningHub said in October 2025 that creators earn cash rewards based on runs, favourites and originality.
Generates images and video with hosted closed and open models such as Seedance 2.5, Vidu Q4 Preview and Nano Banana 2.1, billed per second.
The homepage on October 17, 2026 quoted Nano Banana 2.1 from 0.008 USD a second.
A canvas with built-in AI agents that plan and run multi-step image, video, audio and 3D projects with RunningHub's models and nodes.
Described in a promotional article of May 22, 2026.
A Workflow API, an AI App API, a Model API and an LLM API compatible with OpenAI, Anthropic and Gemini protocols; jobs are submitted, then polled by task ID.
Per the docs, personal member keys can call workflows and AI apps; model and LLM calls need a pay-as-you-go enterprise key.
Open-source plugins on GitHub let a local ComfyUI send work to RunningHub workflows and get images, video, audio and text back.
ComfyUI_RH_APICall had 302 GitHub stars on October 17, 2026.
Signed-in users get 100 RH coins a day, valid for 24 hours, with tasks up to 20 minutes and one task at a time.
At 0.2 coins a second, 100 coins buy about 8 minutes on a 24 GB machine.
The people it helps and the problems it solves, and where it ranks.
Built for
Problems it solves
Graded in
Public tests other people ran: what they gave it and how it did.
Case 1
Input
Run a published workflow in Standard mode (24 GB machine, 0.2 RH coins a second) or Plus mode (48 GB, 0.4 coins a second), paid from the 100 free daily coins or a plan's monthly coins.
What happened
By RunningHub's rates, the 100 free coins cover about 500 seconds in Standard mode, and the Basic plan's 36,000 monthly coins (89.9 USD a year) cover 50 hours. The catch: plan coins expire after 31 days, and payments are not refunded.
Case 2
Input
A local ComfyUI graph that changes the prompt, seed and batch size of a RunningHub text-to-image workflow through a NodeInfoList node and receives the images.
What happened
The plugin sends inputs and returns images, video frames, audio and text, and since October 15, 2025 it falls back to HTTP polling when the connection drops. Users report open bugs on GitHub, including an alpha-channel bug that breaks cut-out workflows, open since May 3, 2025.
Case 3
Input
Submit a Workflow API task with new inputs, receive a task ID, then poll it until the output is ready.
What happened
The docs describe the same submit-and-poll flow for workflows, AI apps and models. Workflow and AI app calls work with a personal member key; plans allow 1 to 5 tasks at once, and the four largest wallet plans cannot call the API at all.
What it does badly, with the evidence, and what we could not settle.
Weak
RunningHub takes broad rights over what you upload and generate.
Evidence: The China site's agreement (June 5, 2026) gives the company a worldwide, perpetual, free, sublicensable licence to your content, including to improve its AI models, unless you opt out by email. The international agreement grants a similar sublicensable licence, including promotional use of your name.
Disputes on both sites go to a Beijing court under Chinese law.
Evidence: The runninghub.ai agreement is with Hong Kong Huize Technology Co., Limited but applies the law of the People's Republic of China and the Chaoyang District court in Beijing; the runninghub.cn agreement does the same with Anhui Haima Cloud.
Coins expire and payments are not refunded.
Evidence: Plan coins last 31 days and daily free coins 24 hours; bought coins last a year. Both agreements say paid services are non-refundable once activated, except for serious defects or where the law requires a refund.
Uploaders carry the copyright risk of shared workflows.
Evidence: Users must own or be licensed for what they upload and bear liability for third-party claims; RunningHub handles infringement complaints within 15 working days and may suspend accounts that do not respond within 2 working days.
Who should not use it now: studios with confidential client material or rules that keep data out of Chinese jurisdiction, and authors who need exclusive control of their workflows.
Evidence: The licences above let RunningHub reuse and sublicense uploads, and even the international site applies Chinese law with a Beijing court.
There is little independent user feedback.
Evidence: G2 listed no reviews of RunningHub on October 17, 2026, and we found no Trustpilot page or Reddit threads; most coverage we found is press releases and promotional articles.
Uncertain
How the companies behind the two sites are related.
Why unsure: runninghub.cn names Anhui Haima Cloud and runninghub.ai names Hong Kong Huize Technology; Haima Cloud is the press contact for runninghub.ai, but neither site explains the link.
Prices on the China site.
Why unsure: The plan page on runninghub.cn requires signing in, which we did not do; its homepage quotes model prices in yuan per second.
Launch date and size.
Why unsure: RunningHub has not published a launch date; the earliest dated trace we found is a GitHub issue on its plugin from December 29, 2024. Its developer counts (50,000 in October 2025, 15,000 active in May 2026) are its own and do not line up.
How creator rewards are paid.
Why unsure: The site links its reward rankings to a Notion page that did not load for us, so we could not confirm how payouts are calculated.
The numbers, each with its source and the date we checked it.
| Spec | Value | Note and source |
|---|---|---|
| Operator in China | Anhui Haima Cloud Intelligent Technology Development Co. (安徽海马云智能科技发展有限公司), Hefei | The site footer lists its Chinese ICP licence, value-added telecom licence and algorithm filing numbers. Source: runninghub.cnchecked |
| Contracting company abroad | Hong Kong Huize Technology Co., Limited; Chinese law, Chaoyang District court in Beijing | Source: runninghub.aichecked |
| Parent company | Haima Cloud (海马云) | Named as media contact on RunningHub's October 24, 2025 release; Haima Cloud's CTO Wen Jian spoke for RunningHub in a March 2025 interview. Source: markets.financialcontent.comchecked |
| Yearly plans (runninghub.ai) | Lite Plus 71.9 USD, Basic 89.9 USD, Basic Plus 152.9 USD, Pro 139 USD, Pro Plus 236.9 USD a year (sale prices); four wallet plans 530 to 1,629 USD a year without API access | Each plan pays out monthly either RH coins (36,000 on Basic) or a USD wallet for closed models. Source: runninghub.aichecked |
| Free allowance | 100 RH coins per daily sign-in, valid 24 hours; tasks up to 20 minutes; 1 task at a time | Source: runninghub.aichecked |
| GPU rates | Lite 0.02 RH coins a second; Standard (24 GB) 0.2; Plus (48 GB) 0.4; 90-series GPUs | Source: runninghub.aichecked |
| Coin expiry and refunds | Plan coins 31 days, bought coins 1 year; no refunds after payment | Source: runninghub.aichecked |
| APIs | Workflow, AI App, Model and LLM APIs; model and LLM calls need a pay-as-you-go enterprise key | Source: runninghub.aichecked |
| Rights over your content (China site) | Worldwide, perpetual, free, sublicensable licence, including to improve AI models | Agreement updated and effective June 5, 2026. Source: runninghub.cnchecked |
| October 2025 price cut | Shared API from 0.7 USD an hour on a 24 GB machine and 0.9 USD on 48 GB; API concurrency raised from 30 to 50 | RunningHub's own announcement. Source: markets.financialcontent.comchecked |
Answer first, then why, an example, the limits and what to do next.
RunningHub is a cloud platform where you build and run ComfyUI workflows, or one-click AI apps made from them, on rented GPUs instead of your own computer. It also hosts closed video and image models such as Seedance 2.5 and Vidu, an agent canvas called rhTV, and APIs for developers.
RunningHub is a Chinese product of Haima Cloud (海马云): the China site is operated by Anhui Haima Cloud Intelligent Technology Development Co. in Hefei. The international site, runninghub.ai, contracts through Hong Kong Huize Technology Co., Limited, but still applies Chinese law, with disputes heard in Beijing.
The China site runninghub.cn is run by Anhui Haima Cloud under a Chinese ICP licence and quotes prices in yuan, while runninghub.ai is the international site priced in US dollars. Each has its own user agreement with a different contracting company; we could not confirm whether accounts or content carry over between them.
RunningHub gives 100 free RH coins per daily sign-in, enough for about 8 minutes on a 24 GB machine, and paid plans on runninghub.ai start at 71.9 USD a year. Workflows cost 0.02 to 0.4 coins a second depending on the machine, plan coins expire after 31 days, and payments are not refunded.
You keep the copyright in what you upload and generate, but RunningHub gets a broad licence to use it. The China site's agreement makes that licence perpetual, free and sublicensable, including to improve its AI models; uploaders also carry the legal risk if a shared workflow infringes someone else's rights.
RunningHub suits people who want to run heavy ComfyUI workflows without a big GPU or publish them as apps; LiblibAI is a broader model community, and local ComfyUI costs nothing if your graphics card is strong enough. LiblibAI also offers online ComfyUI, WebUI and LoRA training, and ComfyUI itself runs on as little as 4 GB of VRAM. Comfy Org's own Comfy Cloud starts at 20 USD a month.
This page is built from RunningHub's two websites, plan page, API docs, user agreements and GitHub plugins, its press releases, an interview with its parent company's CTO, and the alternatives' official pages. Every fact links to its source, and we checked each one on October 17, 2026.
RunningHub belongs to Haima Cloud (海马云), a Chinese cloud company: Haima Cloud is the media contact on RunningHub's October 2025 press release (GetNews), and its CTO Wen Jian spoke for RunningHub in a March 2025 interview (Woshipm). The two websites name different companies.
| runninghub.cn | runninghub.ai | |
|---|---|---|
| Company in the agreement | Anhui Haima Cloud Intelligent Technology Development Co. (安徽海马云智能科技发展有限公司), Hefei | Hong Kong Huize Technology Co., Limited |
| Law and court | Chinese law, Chaoyang District court, Beijing | Chinese law, Chaoyang District court, Beijing |
| Prices | Yuan, per the homepage | US dollars |
| Filings shown | Chinese ICP licence and algorithm filing | None |
Sources: runninghub.cn, China agreement, international agreement, runninghub.ai.
Its own counts keep moving: 50,000 developers in more than 140 countries, 9,000 nodes and 40,000 models in October 2025 (GetNews); 15,000 active developers, 13,681 nodes and more than 100,000 models in May 2026 (China.com). Neither set is independently checked.
Yearly plans on runninghub.ai at the sale prices shown on October 17, 2026; plan names are translated from the Chinese page (plan page):
| Plan | Price a year | Monthly allowance (pick one) | Tasks at once |
|---|---|---|---|
| Free | 0 USD | 100 RH coins per daily sign-in | 1 |
| Lite Plus | 71.9 USD | 13,000 coins + 0.9 USD wallet, or 2,000 coins + 8.9 USD | 1 |
| Basic | 89.9 USD | 36,000 coins + 1 USD, or 4,000 coins + 9.9 USD | 2 |
| Basic Plus | 152.9 USD | 54,000 coins + 1.7 USD, or 8,000 coins + 16.9 USD | 3 |
| Pro | 139 USD | 54,000 coins + 1.5 USD, or 6,000 coins + 14.9 USD | 3 |
| Pro Plus | 236.9 USD | 72,000 coins + 2.6 USD, or 8,000 coins + 25.9 USD | 5 |
Four larger plans (530 to 1,629 USD a year) mostly fund a wallet for closed models and cannot call the API. Running costs 0.02 coins a second in Lite mode, 0.2 on a 24 GB machine and 0.4 on a 48 GB machine. At those rates the 100 free coins buy about 8 minutes on a 24 GB machine, and Basic's 36,000 coins about 50 hours. Free tasks stop at 20 minutes, paid ones at 60. Plan coins expire after 31 days, bought coins after a year, and payments are not refunded.
Both agreements say you keep the rights in what you upload and generate, but you must hold those rights and you carry the liability for claims. RunningHub handles infringement complaints within 15 working days and can suspend an account that does not respond within 2 working days (China agreement, international agreement). The China agreement, dated June 5, 2026, gives RunningHub a worldwide, perpetual, free, sublicensable licence to your content, including to improve its AI models, unless you ask by email to opt out. The international agreement grants a similar sublicensable licence and lets RunningHub use your name in its promotion. Both ban pornography and content that infringes portrait or privacy rights.
Independent feedback is thin: G2 lists no reviews (G2), and we found no Trustpilot page or Reddit threads. On GitHub, ComfyUI_RH_APICall plugin users report an alpha-channel bug that breaks cut-out workflows (open since May 3, 2025) and intermittent list index errors (open since March 4, 2025); most other issues are feature requests (GitHub).
| Alternative | What it is | Price facts |
|---|---|---|
| LiblibAI | Chinese model community with online WebUI, ComfyUI, LoRA training and AI apps | 20 free points a day for signed-in users (LiblibAI) |
| Tensor.Art | Online image generator and model hosting site with a workflow section | Free to start (Tensor.Art) |
| Comfy Cloud | ComfyUI hosted by Comfy Org on 96 GB RTX 6000 Pro cards | From 20 USD a month; runs up to 30 minutes on the lower plans (Comfy Cloud) |
| RunComfy | Cloud ComfyUI billed by the hour | 24 GB machine 1.75 USD an hour, 48 GB 2.50 USD, pay as you go (RunComfy) |
| ComfyUI on your own PC | GPL-3.0 open-source software | Free; runs on as little as 4 GB of VRAM and 8 GB of RAM (GitHub) |
LiblibAI said in July 2024 that it had raised several hundred million yuan within a year and served nearly 10 million creators (TMTPost).
Sources
Changelog
Higgsfield AI is a paid web platform from Higgsfield Inc. that runs more than 30 image and video models from other companies, including Veo 3.1, Kling 3.0 and Seedance, beside its own tools, with a limited free tier and paid credit plans from US$19 a month. In February 2026 it drew public complaints over 'unlimited' plans and creator payouts.
VideoEnhancer is a web tool that upscales short MP4 clips to 1080p, 2K or 4K with the FlashVSR model and also compresses, face-swaps and generates video. After a 5-second free trial, 1080p costs 3 credits a second, $0.03 to $0.09 depending on how credits are bought. VideoEnhancer is our own product: it is made by the team behind GenGrades.
LayerGrab splits a flat image into up to 16 named, transparent layers over a filled-in background, on the web, in Figma or in Chrome, or pulls out only what you name. Each layer costs $0.10 on its $12 plan, $0.08 on Premium. No independent review exists yet. LayerGrab is our own product: it is made by the team behind GenGrades.
VideoTextRemover is a web tool that erases burned-in subtitles, captions and other on-screen text from MP4 and MOV videos and repaints what was behind them, up to 1080p. It bills per second processed, about $0.025 to $0.049, with 30 free seconds a month in some countries. VideoTextRemover is our own product: it is made by the team behind GenGrades.
MagicRemover is a browser tool that erases people, objects, text, logos and watermarks from photos, and subtitles from videos. Photos get 2 free removals a day, 2 more after sign-in, at about 1 megapixel; beyond that, credits start at $9.90 for 330. MagicRemover is our own product: it is made by the team behind GenGrades.
Hermes Agent is the best OpenClaw alternative for most people: free, open source, safer by default and able to import an OpenClaw setup in one command. NanoClaw puts every session in a container, ZeroClaw is a tiny sandboxed binary, Claude Cowork suits people who would rather pay than host, and Hostinger is the cheapest way to keep OpenClaw itself.