cmfy.cloud No Deployment. Just POST.
Run your ComfyUI workflows in the cloud with zero setup. Same JSON format you already use.
# workflow.json - exported from ComfyUI
{
"prompt": {
"1": { "class_type": "CheckpointLoaderSimple", "inputs": {
"ckpt_name": "https://huggingface.co/.../flux1-schnell.safetensors" } },
"2": { "class_type": "CLIPTextEncode", "inputs": { "text": "a sunset...", ... } },
"3": { "class_type": "KSampler", "inputs": { "steps": 20, ... } },
"4": { "class_type": "VAEDecode", "inputs": { "samples": ["3", 0], ... } },
"5": { "class_type": "SaveImage", "inputs": { "images": ["4", 0] } }
}
}
# POST to API
curl -X POST https://api.cmfy.cloud/v1/jobs \
-H "Authorization: Bearer $API_KEY" \
-d @workflow.json
# Response
{ "job_id": "job_abc123", "status": "queued" }From local workflow to production API in 3 steps
cmfy.cloud runs your ComfyUI workflows exactly as you built them. No rewrites, no new SDKs, no deployment hassle.
Use your existing workflows
Export from ComfyUI, POST to our API. Same JSON format you already know.
Production-ready instantly
No servers to provision, no dependencies to install, no infrastructure to manage.
Pay only for compute time
No cold start charges. You pay for generation, not for models loading.
Scales automatically
From 1 to 10,000 requests. We handle load balancing, queuing, and retries.
Here's all it takes
- 1
Export your workflow JSON from ComfyUI
Click "Export (API)" — that's the exact format we use
- 2
Use URLs for models and inputs
Point to HuggingFace, CivitAI, S3, Azure, or any HTTP URL. Add credentials for gated or private models.
- 3
POST to our /v1/jobs endpoint
Get back your generated images. That's it — you're in production.
How It Works
Three steps. No configuration. No learning curve.
Export
Export your workflow JSON from ComfyUI. Use the built-in "Export (API)" option.
Configure
Use URLs for models and inputs. Works with HuggingFace, CivitAI, S3, Azure, or any HTTP URL. Need gated or private models? Add your credentials in the portal.
POST
Send to our /v1/jobs endpoint. Get results via webhook or poll for completion.
{ "prompt": { "1": { "class_type": "CheckpointLoaderSimple", "inputs": { "ckpt_name": "https://huggingface.co/stabilityai/sdxl/resolve/main/sd_xl_base_1.0.safetensors" } }, "2": { "class_type": "CLIPTextEncode", "inputs": { "text": "a beautiful sunset over mountains", "clip": ["1", 1] } }, "3": { "class_type": "CLIPTextEncode", "inputs": { "text": "", "clip": ["1", 1] } }, "4": { "class_type": "KSampler", "inputs": { "model": ["1", 0], "positive": ["2", 0], "negative": ["3", 0], "steps": 20 } }, "5": { "class_type": "VAEDecode", "inputs": { "samples": ["4", 0], "vae": ["1", 2] } }, "6": { "class_type": "SaveImage", "inputs": { "images": ["5", 0] } } }, "webhook_url": "https://your-app.com/webhook"}Everything You Need
Production-ready features that just work. No configuration required.
Native Format
Same JSON you already know from ComfyUI. Zero learning curve, zero format conversion.
Smart Routing
Jobs automatically route to nodes with your models already cached. Minimal cold starts.
Your Models, Your Credentials
Access gated models like Flux with your HuggingFace token. Use private S3 or Azure storage. Credentials encrypted at rest.
Webhook Delivery
Get results pushed to your server. Or poll if you prefer. Your choice.
1100+ Custom Nodes
11 popular node packs pre-installed. IPAdapter, ControlNet, Impact Pack, and more.
Pay Only for What You Use
No subscriptions required. No idle costs. Just pay per second of GPU time.
- No cold start billing
- No minimum commitment
- No hidden fees
Frequently Asked Questions
Everything you need to know to get started.
Any workflow that runs in ComfyUI and fits within 24GB of VRAM and 60GB of RAM. We support the native ComfyUI workflow format, including all standard nodes and 1100+ nodes from popular custom node packs like IPAdapter, ControlNet, Impact Pack, and more.
You only pay for the time ComfyUI takes to execute your prompt—model loading and file downloads are free. Just add credits to your balance and they're deducted as you run jobs. Billing is per-second, rounded up to the nearest second. Failed jobs are not charged. Subscriptions are also available for lower unit costs, higher concurrency, and support SLAs.
Any model that fits in 24GB of VRAM. Load from HuggingFace, CivitAI, or your own storage—including gated and private models. Add your credentials in the portal to access gated models like Flux, or pull from any S3-compatible storage (AWS, Cloudflare R2, Backblaze B2) or Azure. Models are cached after first download for faster subsequent jobs.
Yes! Store your credentials securely in the portal and link them to your API keys. We support HuggingFace tokens for gated models like Flux, credentials for any S3-compatible storage (AWS, Cloudflare R2, Backblaze B2), Azure SAS tokens, and generic HTTP authentication. Your credentials are encrypted at rest and only used during model downloads.
Three options: (1) Use sync mode and wait for the request to complete, getting download links in the response. (2) Provide a webhook URL and get download links in the webhook payload. (3) Poll the job status endpoint and get download links when completed.
We pre-install 11 popular node packs with 1100+ nodes, covering the vast majority of common workflows. This includes IPAdapter, ControlNet preprocessors, Impact Pack, Video Helper Suite, and more. See the full list in our docs.
Jobs with cached models typically start within seconds. Our smart routing system sends jobs to nodes that already have your models loaded, minimizing cold starts.
New accounts receive $1 in free credits—enough for hundreds of images. After that, it's pure pay-as-you-go with no minimums. Add $5 or more to your balance whenever you need to run jobs.
Ready to Ship Your Workflows?
Join the waitlist and get early access with $1 in free credits—enough for hundreds of images.
You're on the list! Check your email for confirmation.
Something went wrong. Please try again.