ComfyUI Online, With Every Node Already Installed
62 custom node packs and 3,300+ nodes are preinstalled and kept current. Open a browser tab and build — no Python environment, no dependency conflicts, no GPU to rent.

Why Run ComfyUI on Diffus
ComfyUI is free and open source — the question is never whether it is good, it is who keeps 62 node packs, 70,000+ models and a GPU running for you.
62 custom node packs ship preinstalled — Impact Pack, KJNodes, WAS Node Suite, Comfyroll, LayerStyle, Easy-Use, Inspire Pack, controlnet_aux and more — on top of ComfyUI's 557 built-ins. No pip, no CUDA mismatch, no broken graph after an update.
55,000+ LoRAs and 10,000+ checkpoints are on disk and selectable from any loader node. Paste a Civitai link to add one that isn't. A local ComfyUI makes you download every one of those yourself.
Flux, Qwen, Wan, LTX-Video, HunyuanVideo, Z-Image, SD 3.5, Nunchaku and GGUF quantisation are already wired up. We track the wrappers so your graph runs the new model instead of waiting on your own install.
Graphs run on our cross-cloud GPU pool. Close the laptop mid-render, come back to the result. No 24 GB card to buy, no hourly instance to remember to shut down.
Drop in a workflow JSON, or a PNG carrying its metadata, and the nodes it references are almost certainly already installed — no red boxes, no hunting down the pack that broke the graph.
Workflows save to your account and reopen on any machine; outputs land in your gallery with their parameters attached instead of in a local folder.
3,300+ Nodes, Nothing to Install
70,000+ Models Already Loaded
New Models the Week They Land
Someone Else Owns the GPU
Someone Else's Workflow Just Runs
Your Graphs Follow You
What the Preinstalled Packs Actually Give You
The node packs below are already there. These are the things people install ComfyUI to do.
Detailing, Upscaling and Face Work
Impact Pack and Impact Subpack for detector-driven face and hand fixes, Ultimate SD Upscale and SUPIR for upscaling, RMBG and Segment Anything for masks. Wired and ready to drop into a graph.

Control Over Composition
controlnet_aux, Advanced-ControlNet, IPAdapter Plus, InstantID and LayerStyle — pose, depth, reference-image identity and layered compositing, all from packs that are already installed and version-matched to the core.

Video, From the Same Graph
WanVideoWrapper, LTXVideo, HunyuanVideoWrapper, AnimateDiff-Evolved, VideoHelperSuite and Frame Interpolation. Send a still into a video graph without leaving the tab or renting a second machine.

How it works
From Zero to a Running Graph
The install step is the one you skip.
ComfyUI loads with all 62 node packs present. Nothing to clone, nothing to pip install.
Import a workflow JSON from anywhere — the nodes it references are almost certainly already here.
70,000+ models are in the loaders. Queue the prompt; our GPUs do the work.
Open the Tab
Load or Build a Graph
Pick a Model and Run
Prefer a Single Button?
Not every job needs a node graph. These run on the same models and the same GPUs.
ComfyUI on Diffus — FAQs
Skip the Install. Keep the Graph.
Open ComfyUI with 62 node packs and 70,000+ models already in place.
No Setup
Cloud GPUs

