RunningHub Review: Features, Pricing, Pros and Cons

ComfyUI is the best image and video tool most people never get working. You clone the repo, install a dozen custom nodes, hit a CUDA version mismatch, and three hours later you are reading a GitHub issue from 2024 written by someone who gave up.

RunningHub's pitch is that you skip all of that. It is a cloud AI content platform that runs ComfyUI workflows on someone else's GPUs, wrapped in one-click apps and a single API covering more than 400 models. Open a workflow, press run, done. So this RunningHub review has one question to answer: is it worth paying for?

The technology is good. The pricing is the problem, and not in the way you would expect. RunningHub will happily tell you what its developer API costs per second. It will not tell you what a subscription costs until you have made an account, and it will not tell you what its in-house currency is worth at all.

The short version is that it depends on how much uncertainty you can stomach. This review covers what the platform does, what is genuinely published about the cost, what is hidden, the refund terms sitting in the service agreement, and who should walk away. All prices were checked on 2 November 2026.

Quick verdict

What works well:

• It removes the single biggest barrier to ComfyUI. No install, no GPU purchase, no dependency archaeology. You open a workflow in a browser tab and run it.

• The model catalogue is current and video-heavy. Roughly 322 video models against 189 image models, and new releases land fast. Nano Banana 2.1, Seedance 2.5, Vidu Q4 Preview and MiniMax H3 are all live.

• The developer API is priced openly, in actual dollars. Per second, per call, per image or per million tokens, all visible at runninghub.ai/call-api with no account needed.

• Billing is per second, not per hour. You pay for compute you use rather than for a machine sitting idle while you think.

• There is a real free allowance. Reported consistently at 100 RH Coins credited on each daily sign-in, which works out to roughly eight minutes of standard compute a day.

• One account covers two different jobs. Hand-built node workflows and one-call access to commercial video models normally live on two separate platforms and two separate invoices.

What to watch:

• You cannot see what a subscription costs before you sign up. The plan page opens a login wall, and both the pricing and subscribe URLs return an HTTP 401.

• RH Coins have no published dollar value. Compute burns 0.2 to 0.4 coins per second, and nothing public states what a coin costs. Platform compute therefore cannot be priced in advance. At all.

• RunningHub's own API Pricing link is empty. The link in its footer leads to a page that displays "No Data".

• Payment is non-refundable by default. Once a service is activated, the service agreement says no refund unless severe defects make it unusable.

• Purchased coins expire after one year, with no refund for whatever is left in the balance.

• Cloud ComfyUI is still ComfyUI. Hosting solves the install problem. It does not solve the node graph, which remains genuinely intimidating if you have never seen one.

• The operator is disclosed only in the fine print. Nothing on the homepage tells you who runs the platform or which courts hear a dispute.

What RunningHub actually is

RunningHub is a cloud platform built around hosted ComfyUI, with one-click apps, an infinite canvas and a 400-plus model API bolted around it. The ComfyUI part is the anchor. Everything else exists so that you are not forced to use it.

If you have never met ComfyUI, think of the difference between a microwave and a kitchen. A text-to-image website is a microwave: one button, one result, very hard to ruin. ComfyUI is a kitchen. You wire the steps together yourself, which means you can make something nobody else can, and also that you can set the place on fire.

Most people want the kitchen and cannot face installing one. That is the gap RunningHub sells into.

Around the core sit four other things. AI Apps are one-click tools for jobs like upscaling, image-to-anime, text-to-video and image-to-video, for when you want a result rather than a graph. rhTV is an infinite canvas for assembling and iterating on generations. Quick Create is the fast path. Explore is a public gallery of other people's workflows and outputs, and it is quietly the most useful part of the platform if you are learning, because a workflow that already works beats any tutorial.

Then there is the Model API, a single standard interface to over 400 models across video, image, audio and 3D, plus endpoints for LLMs, AI applications and whole workflows.

Here is the part worth noticing: video is the larger half of this platform, not image. Roughly 322 video models against 189 image models. If you came expecting another Stable Diffusion front end, you have undersold what this is. You can browse the whole catalogue on RunningHub without an account, and it is worth doing that before you get to the pricing section below.

RunningHub's key features

• Cloud ComfyUI workflows. Run node-based workflows in one click on RunningHub's hardware, with no local install, GPU or dependency setup.

• A video-first model catalogue. Around 322 video models covering text-to-video and image-to-video, including recent releases like Seedance 2.5, Vidu Q4 Preview and MiniMax H3.

• AI Apps. One-click tools for image generation and enhancement, upscaling, image-to-anime, text-to-video and image-to-video.

• rhTV infinite canvas. A freeform workspace for laying out and iterating on generations instead of working one prompt at a time.

• Model API. One interface to 400-plus models, priced openly in dollars, with per-second, per-call, per-image and per-token rates depending on the model.

• Quick Create and Explore. A fast-path creation mode, plus a community gallery of workflows you can copy rather than build from scratch.

• Open-source ComfyUI plugins. Official plugins published on GitHub let you call RunningHub workflows from a local ComfyUI install, which is more openness than most closed platforms bother with.

• Invite rewards. The homepage states that inviting a new user earns 500 RH Coins for you and 500 for them.

How RunningHub performs in practice

Straight answer first: I have not run paid workflows through RunningHub, and I am not going to invent a test I did not do. What follows is reasoned from what the platform documents about its own behaviour, which is enough to tell you where the friction sits.

The GPU tiers tell you most of what you need to know. Standard gives you 24GB of VRAM at 0.2 RH Coins per second. Plus gives you 48GB at 0.4 coins per second. In practice, 24GB is comfortable for image workflows and short video clips, and 48GB is what you reach for when a video model or a long node chain refuses to fit. That is a sensible split, and per-second billing means a two-minute job costs two minutes rather than a rounded-up hour.

The free tier's limits are the real constraint. One task at a time, a 20-minute cap per task, and coins that expire 24 hours after they land. For image work that is genuinely usable. For video it is tight, because video generation is exactly the job that eats both the clock and the coin balance, and you cannot buy your way past the queue on a free account.

The learning curve has not gone anywhere. This is the honest caveat about every cloud ComfyUI service. Moving execution to a server removes the install, the drivers and the hardware bill. It does not teach you what a KSampler does. If node graphs are new to you, start in Explore, copy a workflow that already runs, and change one node at a time.

RunningHub pricing: what is published and what is not

This is where the review earns its keep, so read it slowly. Every figure below was checked on 2 November 2026, and several are time-limited promotions rather than standing rates.

What RunningHub publishes openly

The developer API is priced in plain US dollars at runninghub.ai/call-api, with no account required. Rates vary by model and by unit.

Credit where it is due: this is proper pricing transparency. You can cost a project before you write a line of code.

One small comedy, because it shows how fast the promotions rotate. On the day I checked, the homepage advertised Seedance 2.5 from $0.036 per second, and also carried a banner urging me to lock in current pricing on Seedance 2.0 at $0.052 per second. The older model was the expensive one. If you are treating any of these numbers as fixed, do not.

Platform compute, priced in a currency with no exchange rate

Platform compute does not work like the API. It is billed per second in RH Coins, which is RunningHub's internal currency.

Those are real published numbers. They are also, on their own, completely useless.

Because nothing public tells you what an RH Coin is worth. Not the homepage, not the app pages, not the legal documents. So 0.2 coins per second converts into precisely nothing. A one-minute Standard render costs 12 coins, and 12 coins costs an unknown amount of money.

I could guess. Several tool directories have guessed, and they disagree with each other by a factor of four, which tells you roughly what their guesses are worth. We are not adding to that pile. The conversion rate obviously exists. It just is not for you until you have handed over an email address.

What RunningHub does not publish

That fourth row is the one that bothers me most. RunningHub's own footer carries a link for API pricing, and the page it leads to is empty. Nobody checked. And the opacity is harder to forgive precisely because the developer API is priced so well, which proves this is a choice rather than a limitation.

The free tier

RunningHub appears to credit 100 RH Coins on each daily sign-in, valid for 24 hours, with one task at a time and a 20-minute cap. At the Standard burn rate that is roughly eight minutes of compute a day, which is enough to learn the platform and finish small image jobs.

Be clear about the status of that figure. It is reported consistently by multiple independent sources and in community posts on RunningHub's own domain, but it does not appear in any official documentation reachable without an account. Treat it as reported rather than confirmed.

The fine print, which is worse than the opacity

The Paid Service Agreement is where guessing gets expensive. Three clauses matter:

• Non-refundable once activated. Payment is not refundable after the service starts, unless severe defects make it unusable.

• Cancel early, get nothing back. Ending a plan before the term expires earns no refund for the remaining period.

• Coins expire. Purchased RH Coins are valid for one year, after which the agreement states no refund is given. They also cannot be exchanged for cash or transferred to anyone else.

Put those next to the hidden prices and you have the actual problem with RunningHub. You cannot see the price, you cannot convert the currency, and once you have paid you cannot get it back. Any one of those is survivable. All three together means your first purchase is a bet.

Signing up through our link

If you register through our RunningHub link, there is a sweetener. RunningHub's homepage states that an invite grants RH Coins to both parties, quoted publicly at 500 each, and AIwithKay readers get additional signup credits on top under our arrangement with RunningHub.

We cannot independently verify the exact figure, because RunningHub publishes no referral terms to anyone who is not logged in. So check the balance actually credited to your account on signup, and do that before you buy anything.

Who runs RunningHub

This is not on the homepage, and for some readers it will matter more than the price does.

The legal documents name the operator as Hong Kong Huize Technology Co., Limited. The privacy policy was last updated on 10 March 2025 and refers to data being shared as necessary within a "Haima Cloud ecosystem". The user agreement is governed by the laws of the People's Republic of China, and disputes go to the People's Court of Chaoyang District, Beijing.

None of that is a scandal, and plenty of excellent AI infrastructure is built exactly this way. It is simply a fact you should know before you upload unreleased client work, because it defines where your files sit and which court you would be standing in if something went wrong. If you work under contracts with data residency clauses, check them against this first. If you are making anime wallpapers for fun, it changes nothing.

RunningHub vs the alternatives

RunComfy is the closest direct competitor and the sharpest contrast. Its pricing page lists every GPU tier by the hour, from $0.50 upward, billed per second, with no login required. You can budget a month before you create an account. RunningHub beats it on model breadth and on the aggregation API. RunComfy beats it on letting you do arithmetic.

ThinkDiffusion is the gentlest option, with published monthly plans from $19.99 and a deliberately simple session model. Less power, and far less confusion.

Replicate and fal.ai overlap with RunningHub's API rather than its ComfyUI side. Both publish per-model rates openly. Neither gives you a hosted node editor, so if the workflow canvas is why you are here, they are not substitutes.

And if you only want finished video and have no interest in nodes at all, go to the source rather than the aggregator. Our Kling AI review and Luma AI review cover the two most popular single-app generators, Higgsfield AI is the one to read if cinematic camera control is what you care about, and our Seedance vs Veo comparison covers one of the exact models RunningHub resells. Every one of them publishes its prices.

Who should use RunningHub, and who should skip it

Use it if:

• You already know ComfyUI and are tired of maintaining a local install, the drivers and a GPU.

• You want one API key across 400-plus video, image, audio and 3D models instead of five vendor accounts.

• Video is your main output and model breadth matters more to you than hand-holding.

• You want to experiment on the free daily allowance without ever entering a card.

• You can tolerate spending a small amount first in order to find out what things cost.

Skip it if:

• You need to budget before you commit. You cannot, and no amount of goodwill fixes that.

• You need a refund path. Non-refundable once activated is the default here, not the exception.

• You have never used a node graph and want something that works on day one. Start with a single-app generator instead.

• Your work falls under a contract with data residency or jurisdiction requirements.

• You are buying for a team and need an invoice with a predictable number on it.

The verdict

RunningHub scores 6.5 out of 10 on our Kay Score, and the entire gap between what it is and what it scores is about money.

The engineering deserves better. Cloud ComfyUI execution that simply works, 322 video models kept genuinely current, an infinite canvas, and an aggregation API priced in the open with per-second granularity. On capability alone this is an 8-plus platform, and the open-source plugin work on GitHub suggests a team that actually cares about the ecosystem it sits in.

What pulls the score down is not a missing feature. It is that you cannot see the subscription price, you cannot convert the currency into money, and you cannot get a refund by default. Three compounding unknowns, arriving at the exact moment you are asked to pay, from a company that has already proved on its API page that it knows how to publish a price properly.

So the recommendation is conditional, and I will state it plainly. If you know ComfyUI, want the hardware problem to go away, and can treat your first purchase as tuition, try RunningHub on the free daily allowance first and only buy once you can see the rates inside your own account. If you need a number you can put in a budget this week, use RunComfy or ThinkDiffusion and come back when RunningHub publishes a price list.

Fix the pricing page and this becomes an easy recommendation. Right now it is a good platform asking for an unusual amount of trust up front.