--- title: "Comfy Cloud Nodes: Run Open Models on Our GPUs, From Your ComfyUI" description: "Drop a Comfy Cloud node into a local graph and run open models on Comfy Cloud GPUs. No subscription, no downloads, billed per GPU-second in credits." canonical: https://comfy.org/cloud-nodes/ lang: en index: https://comfy.org/llms.txt --- CLOUD NODES BETA # Your graph. Our GPUs. Comfy Cloud nodes run the newest open models on our GPUs from inside your own ComfyUI. Drop one into the graph you already have and wire it like any other node. - No subscription - Nothing to download, no VRAM floor - Billed per GPU-second in credits [Get Started](https://comfy.org/download) [Read the Docs](https://docs.comfy.org/cloud-nodes/overview) ## Four steps to your first run 01 ### Update ComfyUI Update to v0.34.5 or later. The nodes are part of ComfyUI core, so there is nothing to install separately. 02 ### Restart ComfyUI The nodes only register at startup, so they will not appear until you restart. 03 ### Find a node Search the node library for “Comfy Cloud”. 04 ### Wire it in and run Wire it like any other node. The model runs on our GPUs and the result comes back into the same canvas. [How to Update ComfyUI](https://docs.comfy.org/installation/update_comfyui) ## Eight nodes, six models Image, video and audio at launch. Every image and clip below is a real output from the model on the card, generated on Comfy Cloud. - ![Flux 2](https://media.comfy.org/website/cloud-nodes/models/flux-2.webp) ### Flux 2 1 node - ### MiniMax H3 3 nodes - ![Z-Image Turbo](https://media.comfy.org/website/cloud-nodes/models/z-image-turbo.webp) ### Z-Image Turbo 1 node - ![Mage Flow](https://media.comfy.org/website/cloud-nodes/models/mage-flow.webp) ### Mage Flow 1 node - ![Mage Flow Turbo](https://media.comfy.org/website/cloud-nodes/models/mage-flow-turbo.webp) ### Mage Flow Turbo 1 node - ![MiniMax Music 3](https://media.comfy.org/website/cloud-nodes/models/minimax-music-3.webp) ### MiniMax Music 3 1 node The full node list and their inputs live in the docs. [See the node reference](https://docs.comfy.org/cloud-nodes/overview) ## What actually runs on our side ### One curated workflow per node, pinned to a fixed model version. - Pinned versions, so a saved graph keeps its result. - Only the node’s inputs leave your machine. - Results return as a normal IMAGE, VIDEO or AUDIO output. - Charged by execution time, not per generation. ## Maximum control. Offload only the compute. Your graph, your nodes, your outputs. A Comfy Cloud node hands off a single model run and hands the result straight back. ### Your workflow stays yours The graph you built, the nodes around it and the outputs you route onward all stay on your machine, exactly as they are. ### Only the compute moves A Cloud node sends the inputs for one model and returns the result. Everything else in the workflow keeps running locally. ### Reach for any model The newest open models run on our GPUs at full size, so what your hardware can do stops deciding what you can make. ### One node, not a migration You add a node to the graph you already have. Nothing else about your setup changes. Products ## The AI creation engine for complete control Over every model, every node, every step, every output. [Comfy Desktop Run ComfyUI on your own hardware. SEE DESKTOP FEATURES](https://comfy.org/download) [Comfy Cloud The full power of ComfyUI from anywhere. SEE CLOUD FEATURES](https://comfy.org/cloud) [Developer Platform Deploy and scale ComfyUI workflows. SEE PLATFORM FEATURES](https://comfy.org/platform) [Comfy Enterprise Enterprise-grade infrastructure for the creative engine inside your organization. SEE ENTERPRISE FEATURES](https://comfy.org/enterprise) ## Frequently asked questions Because the curated set is still changing: nodes may gain or lose options, and a workflow may be retired. It is not a statement about reliability, runs are real and are billed the same as any other partner node. No. These behave like every other partner node: an account and credits are enough. There is no plan floor. Eight nodes at launch. Four text-to-image (Flux 2, Mage Flow, Mage Flow Turbo, Z-Image Turbo), three MiniMax H3 video nodes (text, image, and first-last frame to video), and MiniMax Music 3 for audio. A curated workflow per node, pinned to an exact model version so a saved graph keeps producing the same thing. Your graph, prompts and inputs stay on your machine apart from what the node sends. Per GPU-second in credits, shown on the node before you run. A short image generation is a few seconds; video is longer. Not yet. Each node exposes a fixed set of weights we host. Bringing your own is the obvious next step and is not in this first release. No, and it should not. Keep running what your machine handles well. These nodes exist for the models it cannot, without asking you to leave your workflow. ## Run the model, not the hardware Update to v0.34.5 or later, search the node library for Comfy Cloud, and wire one in. [Get Started](https://comfy.org/download) [Read the Docs](https://docs.comfy.org/cloud-nodes/overview)