ComfyUI
Use Atlas Cloud models inside ComfyUI with drop-in nodes. No local GPU required — Seedance, Kling, Wan, Veo, Nano Banana, Seedream, FLUX, and more.
The Atlas Cloud ComfyUI nodes let you call hosted models from a ComfyUI graph. Generation runs on Atlas Cloud, so you can use large video and image models without a local GPU — and mix them with the local nodes you already have.
GitHub repository: AtlasCloudAI/atlascloud_comfyui
Requirements
- ComfyUI (desktop app or source install)
- Python 3.10 or newer
- An Atlas Cloud API key — create one at console/api-keys
Installation
cd ~/Documents/ComfyUI/custom_nodes
git clone https://github.com/AtlasCloudAI/atlascloud_comfyui.git
cd atlascloud_comfyui
~/Documents/ComfyUI/.venv/bin/python -m pip install -r requirements.txtThen quit ComfyUI completely and reopen it.
Dependencies are light — requests, numpy, and pillow.
Install the dependencies into the same Python environment ComfyUI runs on. Using a different interpreter is the most common cause of "module not found" after installation.
After restarting, the nodes appear under Node Library → AtlasCloud.
Configuring your API key
The key is not read from an environment variable. Add an AtlasCloud Client node to your graph and paste the key into it — that node holds the credential and the base URL, and every other Atlas Cloud node takes its connection from there.
A saved workflow JSON can contain whatever you typed into the client node. Clear the field before sharing a workflow file publicly.
Quick start
Install the nodes and restart ComfyUI.
Drag examples/01-text-to-image.json from the cloned repository onto the canvas.
Select the AtlasCloud Client node and paste your API key.
Run the graph.
Example workflows
The repository ships runnable graphs in examples/:
| File | What it does |
|---|---|
01-text-to-image.json | Text-to-image, 3 nodes |
02-image-to-video.json | Image-to-video, 5 nodes |
03-multi-reference-video.json | Reference-driven video with up to 8 reference images, 7 nodes |
parallel_video/3x_parallel_t2v.workflow.json | Three text-to-video jobs running in parallel |
Available models
The node set tracks the Atlas Cloud catalog and covers the major families:
- Video — Seedance, Kling, Wan, Veo, Hailuo, Vidu, PixVerse, LTX
- Image — Nano Banana, Seedream, GPT Image, FLUX, Qwen Image, Grok Imagine
- Studio workflows — packaged multi-step pipelines such as product visuals and virtual try-on
Common utility nodes include Image Previewer and Video Previewer for inspecting results inline.
Browse the full catalog in the Model Library.
Optional environment variables
| Variable | Purpose |
|---|---|
ATLAS_ALLOW_DEPRECATED_MODELS | Set to 1 to keep legacy nodes visible after a model is retired, so old workflows still load |
ATLASCLOUD_HISTORY_DIR | Override where generation history is stored locally. Defaults to an application-support directory under your user profile |
Troubleshooting
Related
Last updated on