RunComfy
RunComfy is a cloud platform that runs ComfyUI in your browser with no setup, on dedicated GPUs from 16 to 141 GB of VRAM billed by the second, and turns any saved workflow into an autoscaling serverless API.
What is RunComfy?
RunComfy is a cloud platform built around ComfyUI, the open-source node-based interface used to drive Stable Diffusion and the current generation of image, video and audio diffusion models. Instead of installing ComfyUI, CUDA drivers and gigabytes of checkpoints on a local machine, users open a browser session that already has everything in place. The platform serves ComfyUI unmodified and synced with official releases, so graphs built locally behave the same way in the cloud. Work happens on a dedicated GPU machine rented by the second. Eight tiers are offered, from a CPU-only box for uploads to a T4 or A4000 with 16 GB of VRAM, an A10G or A5000 with 24 GB, an A6000 or L40S with 48 GB, an A100 or H100 with 80 GB, and an H200 with 141 GB, which covers the heaviest current models. More than 200 curated workflow templates span video generation and transformation, image generation and editing, restoration and upscaling, 3D work and music. Two features set the service apart from plain GPU rental. The first is the auto-setup agent: dropping a workflow.json file makes the platform resolve every missing custom node, dependency and model, including gated Hugging Face and CivitAI files, and hand back a link that simply runs. The second is Cloud Save, which freezes the whole environment, graph, custom nodes, models, Python runtime and operating system, into one reproducible snapshot that reopens identically and can be shared through a single link. From there, any saved workflow becomes a production endpoint in one click. The serverless API gives a stable HTTPS URL and a deployment identifier, scales to zero when idle, versions deployments with progressive rollback, and reports queue times and cold starts in a live dashboard. Alongside it sit a Trainer for LoRA fine-tuning, a catalogue of hosted models billed per generation, an MCP server for Claude, Cursor and Windsurf, and a command-line tool. The service is published by Inceptions AI INC, based in Las Vegas, Nevada.
What it does
- Launch a fully configured ComfyUI session in a browser within seconds, with no local install
- Drop a workflow.json file and let the auto-setup agent install every missing custom node and model
- Deploy any saved workflow as an autoscaling serverless API endpoint in one click
- Rent dedicated GPU machines from 16 to 141 GB of VRAM, billed by the second
- Train LoRA adapters on H100, H200 or 8-way H100 machines with persistent checkpoints
- Save a complete environment to the cloud and share it with a colleague or a class through one link
- Call hosted image, video and audio models per generation, or from Claude, Cursor and Windsurf through MCP
When to use RunComfy / When not to
A quick filter to help you decide if RunComfy is the right fit.
When to use RunComfy
- Digital artists and VFX creators who want ComfyUI without managing CUDA, Docker or a local GPU
- Anyone who downloads community workflows from CivitAI, Reddit or Discord and keeps hitting missing nodes and missing models
- Developers who need a reproducible, autoscaling image and video generation endpoint behind their product
- Trainers, schools and public AI programmes that need one identical environment for every student, with nothing to install
- Teams working with heavy models that need up to 141 GB of VRAM and want to hand a colleague the exact same setup through a link
When not to use RunComfy
- People looking for a one-click image generator: ComfyUI is still a node graph and its logic has to be learned
- Buyers who need a fixed, predictable monthly bill, since everything is metered GPU time and Pro only lowers the rates
- Organisations that require data residency in the European Union, as the declared servers are all in the United States
- Procurement teams that need a signed data processing agreement or a named subprocessor list, neither of which is published
- Users who want to work offline or on their own hardware, or who expect a free permanent GPU rather than a shared public server
How to use RunComfy
A typical end-to-end flow, from setup to results.
- Try the service without an account through Free ComfyUI Online, which runs on a shared public server with a common queue, no persistent storage and no support
- Create an account, then either pick one of the 200+ ready-to-run workflow templates or upload your own workflow.json file
- Let the auto-setup agent resolve the missing custom nodes and models, adding a Hugging Face or CivitAI token once if a file is gated
- Choose a GPU tier sized to your models; billing starts by the second as soon as the machine boots, so shut it down when you stop
- Install extra nodes through the preinstalled ComfyUI Manager and pull models from Civitai, Hugging Face or Google Drive
- Upload your own images and videos, folder structures included, through the file browser
- Use Cloud Save to freeze the full environment, then reopen it later or share the link with a colleague, a student or a client
- For production, export the graph through Workflow then Export (API), pick the inputs you want to override at runtime and save to the cloud again
- Deploy that snapshot as a serverless endpoint, set the GPU tier and the autoscaling rules, and collect your deployment identifier
- Call the endpoint asynchronously with your API key, poll for status or use webhooks, and follow spend and cold starts in the dashboard
Pros & Cons
Pros
- ComfyUI is served unmodified, so local and community workflows run as they are and nothing has to be rewritten for the cloud
- The auto-setup agent removes the single most common failure in the ComfyUI world: red nodes, missing models and import-failed errors
- The GPU range is unusually wide for a consumer-facing service, reaching 141 GB of VRAM on an H200
- Per-second billing and scale-to-zero mean idle deployments cost nothing, and one shared balance covers ComfyUI, Trainer and Models
- The same artefact carries a project from prototype to production: a saved workflow becomes an API endpoint without changing tools
- Reproducibility is genuine, since Cloud Save freezes the runtime and operating system rather than just the graph
- Ownership of generated images is contractually handed to the customer, with the publisher waiving all rights over them
Cons
- Spending is hard to forecast, because everything depends on GPU time and the top tier reaches USD 9.59 per hour on pay-as-you-go
- Pro grants no usage quota, so the balance still has to be topped up, and purchased funds expire 365 days after the purchase date
- On pay-as-you-go, storage is capped at 10 GB and inactive assets are deleted after 90 days
- Data is hosted in the United States only, with no region choice, no published DPA, no named subprocessor list and no Article 27 EU representative
- The terms and the privacy policy are Termly iframes rendered in JavaScript, which makes them hard to read, quote or archive
- Corporate transparency is thin: no about page, no contact page, no named team, and a support promise of 24/7 with no published response commitment
- The comparison table on the free ComfyUI page advertises a Free Trial that neither the pricing page nor the terms describe
Pricing & Plans
RunComfy offers a permanently free option: Free ComfyUI Online runs in the browser with no account and no payment card, on a shared public server with no persistent storage and no support. Paid use is pay-as-you-go, billed by the second against a single prepaid balance shared by ComfyUI, Trainer and Models, with GPU rates running from USD 0.99 per hour on a 16 GB machine to USD 9.59 per hour on a 141 GB H200. The only subscription is Pro, whose lowest price point is USD 19.99 per month when billed annually at USD 239.90 per year; paid month by month, Pro costs USD 29.99. Pro cuts every GPU rate by at least 20 percent and adds 20 CPU hours and USD 10 of credits each month, but it does not replace the balance.
- USD 0
- no account and no payment card
- shared public server with a common queue
- no persistent storage
- no support
- no subscription
- standard GPU rates billed by the second
- 10 GB of storage with inactive assets removed after 90 days
- up to 10 saved workflow environments
- serverless API deployment included
- Discord and email support
- USD 29.99 per month or USD 239.90 per year (USD 19.99 per month
- advertised as 33 percent off with four months free)
- at least 20 percent off every ComfyUI and Trainer GPU rate
- 20 CPU hours per month included
- USD 10 of monthly credits
- unlimited saved environments
- 200 GB of storage with permanent retention
- priority support
- no public price
- arranged by contacting hi@runcomfy.com
- the publisher reports collaborations with educational institutions
- government-led AI training programmes and companies
Data, GDPR & hosting
A consolidated view of how RunComfy handles your data.
GDPR overview
The word GDPR does not appear anywhere in the site's readable HTML. Both legal pages are Termly documents embedded in JavaScript iframes, so their text cannot be read without a browser. The underlying document configuration shows a policy that does collect personal data from EU residents and that covers international transfers with the European Commission's Standard Contractual Clauses, although no link to those clauses is published. Two absences are explicit: no EU or UK representative is designated under Article 27, and no data protection officer is appointed. A single address, admin@inceptions.ai, handles every privacy request, alongside a Termly data subject access request form. Named US state laws are covered in more detail than European ones. Treat the European posture as declared but unverified.
Who owns the data?
The Generation & Model Usage Terms state plainly that every image generated on a rented server is the property of the customer, and that the publisher releases all rights to it, will not use it, will not sell it and claims no ownership. Uploaded models, LoRAs and custom nodes stay in the customer's private account storage. Responsibility follows ownership: the publisher accepts no liability for user-generated content and cannot grant commercial clearance, so verifying each model's licence is the customer's job. On the infrastructure side, RunComfy says sessions and API deployments run in dedicated isolated containers that no third party, itself included, can access unless the customer chooses to share them.
Reuse rights
Outputs can be reused freely: the terms transfer ownership of generated images to the customer and the publisher waives every right over them, so there is no permission to request before publishing, selling or modifying a result. The limit is licensing rather than platform policy. The terms explicitly say RunComfy cannot clear anything for commercial use, that individual models carry their own licences, and that the customer should consult the model creators or a copyright attorney. The Stable Diffusion licence is bound into the terms by reference. Saved environments, uploaded models and workflow snapshots remain the customer's, and can be shared with anyone through a link at the customer's discretion.
Data retention & training
Hosting summary
Hosting is declared entirely in the United States. The privacy policy configuration lists the United States as the location of the publisher's servers and of its third-party providers' servers, and the terms of service give the United States as the website hosting country. No European, British or Asia-Pacific region is offered, and there is no way to choose where a workload runs. International transfers are acknowledged and covered by the European Commission's Standard Contractual Clauses, though no copy or link to those clauses is published. The domain resolves to 76.76.21.21, an anycast address geolocated in the United States on Amazon infrastructure, which reflects the front-end rather than where GPU jobs execute. On isolation, RunComfy states that each ComfyUI session and each API deployment runs in its own dedicated container, with no shared processes or memory, and that no third party, including RunComfy itself, can reach that environment unless the customer chooses to share it. Governing law and arbitration are in Nevada.
Things to keep in mind
Risks and trade-offs to weigh before adopting RunComfy.
- The meter runs while the machine is on, not while you are generating: forgetting an idle ComfyUI session open is the fastest way to burn a balance
- Purchased funds expire 365 days after purchase, and on pay-as-you-go inactive assets are deleted after 90 days, so money and work can both quietly disappear
- Prices move: the site itself notes that H200 rates were revised on 1 June 2026, and model rates are presented only as sample rates
- The publisher explicitly refuses to clear anything for commercial use, so licence checks on every model you load remain your legal responsibility
- A tool this frictionless makes it easy to generate faces, voices and likenesses that belong to real people; the character-swap and voice-cloning workflows carry obvious consent and deepfake risks
- Hosting is United States only, with no DPA, no named subprocessor list and no EU representative, which is a real constraint for European or regulated work
- Leaning on ready-made templates and an auto-setup agent can leave you unable to diagnose a broken graph yourself, and the underlying ComfyUI skills atrophy quickly
Setup & Integrations
Technical difficulty
Getting started is easy: nothing to install, no Docker, no CUDA, no Kubernetes, and a free browser session available without an account. The auto-setup agent absorbs the hardest part by installing missing nodes and models on its own. Two things still require judgement. ComfyUI remains a node graph, so checkpoints, CLIP, VAE and samplers have to be understood. And picking a GPU tier means knowing how large your models are; the site suggests starting around 48 GB. On the API side, exporting a graph and calling an asynchronous REST endpoint is within reach of any developer, with no DevOps skills needed.
Deployment
Integrations
Supported languages
Behind RunComfy
Social
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
Do I need to install anything to use RunComfy?
What happens when a downloaded workflow has missing nodes or missing models?
Which GPUs are available and how am I charged?
Do my models and outputs survive when the machine shuts down?
Can I turn a workflow into an API?
Does RunComfy use my generations or my data?
Where is my data hosted?
Is Pro mandatory, and what does the free option really include?
Should you pick RunComfy?
RunComfy is one of the most complete answers available to a very specific frustration: ComfyUI is powerful, and installing it is miserable. The platform removes that friction without watering the tool down, since it serves ComfyUI unmodified and keeps it in step with official releases. The auto-setup agent goes further than most competitors by resolving missing nodes and models from an uploaded graph, and Cloud Save turns an environment into a reproducible artefact rather than a fragile local install. The path from a first experiment to a production endpoint stays inside a single product, which is rare in this space. The GPU range, up to 141 GB of VRAM, and per-second billing with scale-to-zero make heavy work affordable when it is intermittent. The contractual position on outputs is unusually clear and favourable: generated images belong to the customer and the publisher waives every right over them. The reservations are mostly about predictability and transparency rather than about the product. Costs are entirely usage-driven, the Pro subscription discounts rates without granting quota, purchased funds expire after a year, and inactive assets disappear after 90 days on the free-of-subscription tier. On the governance side the picture is thinner: hosting is United States only, there is no published data processing agreement, no named subprocessor list, no Article 27 EU representative and no data protection officer, and the legal documents are JavaScript-rendered embeds that resist reading and archiving. No about page, no contact page and no named team complete a fairly opaque corporate profile. For individual creators, studios and developers building generative media products, RunComfy is a strong and pragmatic choice. For regulated buyers or European organisations with data residency obligations, the gaps deserve a careful look before committing.
- Choosing a selection results in a full page refresh.
- Opens in a new window.