ComfyUI node for cloud generation
Run cloud AI image, video and audio models inside ComfyUI with one custom node file and the PopNow API. No GPU or model downloads; install steps, inputs, pricing and error fixes.
One small custom node puts every PopNow image, video and audio model into a ComfyUI graph. It sends the job to the cloud, waits, and hands the result to the next node: an IMAGE for image models, a saved file for video and audio. Nothing is downloaded to your GPU, so video models that need far more memory than a laptop has run fine from a laptop.
This page matches the ComfyUI page on popnow.ai, which fills in the node's defaults for the model you pick.
Install in four steps
Save the node file
Save the code below as custom_nodes/popnow_api.py inside your ComfyUI folder. It uses only requests, Pillow, NumPy and PyTorch, which ComfyUI already ships.
Set your API key
Create a key in the API console and set it in the environment you start ComfyUI from:
export POPNOW_API_KEY="sk-popnow-..."
python main.pyOn Windows portable builds, add set POPNOW_API_KEY=sk-popnow-... to the .bat file that starts ComfyUI.
Restart and add the node
Restart ComfyUI, double-click the canvas, search for PopNow and add PopNow Generate.
Queue
Type a model ID and a prompt, connect the image output to a Preview or Save node, and queue the prompt.
The node
Read it before you run it; it is short on purpose.
Inputs and outputs
| Input | What to put in it |
|---|---|
model | A model ID from Models, for example google/nano-banana-2.1/text-to-image or minimax/h3/image-to-video |
prompt | What to make. Leave it empty for models that take no prompt |
options | The model's other fields as JSON, such as {"resolution": "2k"} or {"duration_seconds": 5}. Each model's page lists them |
image (optional) | An IMAGE from another node. It is uploaded for you and placed in the field named by image_field |
image_field | Which field gets the image: image_url for most video models, image_urls for image editing |
| Output | Contains |
|---|---|
image | The result as an IMAGE for image models; a blank placeholder for video and audio |
url | The signed file URL, valid for about 20 minutes |
path | For video and audio: where the file was saved in ComfyUI's output folder |
Common setups
| Goal | model | options | image_field |
|---|---|---|---|
| Text to image | google/nano-banana-2.1/text-to-image | {"aspect_ratio": "16:9", "resolution": "1k"} | — |
| Edit a local render | google/nano-banana-2.1/edit | {"resolution": "2k"} | image_urls |
| Animate a still | minimax/h3/image-to-video | {"duration_seconds": 5, "resolution": "768p"} | image_url |
| Upscale a frame | popnow/image-upscaler-v1/upscale | {} | image_url |
| Cut out the subject | popnow/background-remover-v1/edit | {} | image_url |
A typical hybrid graph: make a start frame with a local checkpoint, animate it in the cloud with an image-to-video model, then load the saved MP4 back with a video loader node and upscale or interpolate it locally.
What the node does under the hood
The node is a thin wrapper over three REST calls; the same calls in other languages:
curl -X POST "https://api.popnow.ai/api/v1/jobs/createTask" \
-H "Authorization: Bearer $POPNOW_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: $(uuidgen)" \
-d '{"model": "google/nano-banana-2.1/text-to-image", "input": {"prompt": "A lighthouse on a cliff at dusk, warm window light", "aspect_ratio": "16:9", "resolution": "1k"}}'
curl "https://api.popnow.ai/api/v1/jobs/recordInfo?taskId=job_..." \
-H "Authorization: Bearer $POPNOW_API_KEY"const API = 'https://api.popnow.ai/api/v1';
const auth = { Authorization: `Bearer ${process.env.POPNOW_API_KEY}` };
const created = await fetch(`${API}/jobs/createTask`, {
method: 'POST',
headers: { ...auth, 'Content-Type': 'application/json', 'Idempotency-Key': crypto.randomUUID() },
body: JSON.stringify({
model: 'google/nano-banana-2.1/text-to-image',
input: { prompt: 'A lighthouse on a cliff at dusk, warm window light', aspect_ratio: '16:9', resolution: '1k' },
}),
}).then((r) => r.json());
console.log(created.data.taskId);Uploading the input image is the three-step flow on Media; the result is a popnow://f/… reference that only works with keys on your own account.
Pricing
Each run costs the model's normal price in credits; the node adds nothing. Nano Banana 2.1 at 1K is 8 credits per image, MiniMax H3 at 768p is 28 credits for 5 seconds. Failed runs are not charged. Prices for every model are on its page and in Credits and refunds; use jobs/quote for an exact figure.
Common errors
| ComfyUI shows | Why | Fix |
|---|---|---|
Set POPNOW_API_KEY before starting ComfyUI | The variable is not in ComfyUI's environment | Set it in the same shell or .bat file, then restart |
PopNow 401: … | Key missing, mistyped or revoked | Create a new key in the API console |
PopNow 400: … naming a field | options has a field the model does not accept, or a value out of range | Check the model's page; unknown fields are rejected |
PopNow 40201: … | Not enough credits | Top up, or lower the resolution or length |
PopNow failed: … | The model could not finish or declined the request | Nothing was charged; change the prompt or settings |
JSONDecodeError | options is not valid JSON | Use double quotes: {"resolution": "2k"} |
| Node not in the search list | The file is not in custom_nodes, or ComfyUI was not restarted | Check the path and restart; look for import errors in the console |
Is this an official package on the ComfyUI registry?
No. It is a single file published here and on popnow.ai. Copy it, read it, and update it from this page when the API changes.
Does it work with ComfyUI Desktop and portable builds?
Yes. It works anywhere you can add a custom node and set an environment variable: portable, Desktop and self-hosted servers.
Do I need a GPU?
No. Generation runs on PopNow's side; your machine only sends the request and receives the file.
How do I get video frames back into the graph?
The node saves the video to ComfyUI's output folder and returns its path. Load it with a video loader node to get frames.