ComfyUI を API として実 GPU 上で動かす
Deploy your ComfyUI worker container, get a stable POST /v1/gpu/run/{slug} endpoint backed by autoscaling GPUs. Per-second billing. SDXL, Flux, video — your graph, your nodes, your weights.
From Docker image to live endpoint
curl https://hypereal.cloud/v1/deployments \
-H "Authorization: Bearer $HYPEREAL_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"slug": "my-comfy-workflow",
"name": "My ComfyUI Workflow",
"dockerImage": "runpod/worker-comfyui:dev-cuda12.1.1",
"gpuPoolTier": "ADA_24",
"workersMin": 0,
"workersMax": 5,
"idleTimeoutSeconds": 30
}'curl https://hypereal.cloud/v1/gpu/run/my-comfy-workflow \
-H "Authorization: Bearer $HYPEREAL_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"input": {
"prompt": "cinematic portrait of an astronaut, 35mm film",
"seed": 42,
"workflow_overrides": {
"6.inputs.text": "cinematic portrait of an astronaut",
"3.inputs.seed": 42
}
}
}'
# Async response: { "job_id": "job_01HX...", "status": "queued" }
# Poll GET /v1/gpu/run/my-comfy-workflow/job_01HX...
# Or pass "sync": true for an inline response (up to 240s).Same endpoint shape whether you POST sync (up to 240 s inline) or async (immediate `job_id`, poll or webhook for result).
Why ComfyUI users pick Hypereal
Your custom nodes work
Bring any ComfyUI image — custom nodes, custom samplers, custom checkpoints. We do not rewrite your graph or restrict which nodes you can install. If it runs locally in ComfyUI, it runs here.
Per-second billing, scale-to-zero
You are charged for GPU-seconds while your handler is executing. Workers scale to zero after the idle timeout, so a slow Sunday costs you nothing. Cold-start is < 8 s on warm pool.
Webhook delivery on terminal status
Subscribe a URL to receive a signed POST when a job completes or fails — full input/output payload, retries with exponential backoff, dead-letter on permanent failure. No polling required.
ワークフロー
- 01
Build your ComfyUI image
Start from the published `runpod/worker-comfyui` base or roll your own. Bake in your nodes, your checkpoints, and a `handler.py` that loads the graph and exposes `runpod.serverless.start`.
- 02
Push to GHCR or Docker Hub
Any public OCI image works out of the box. Private registries are supported via image-pull credentials configured per workspace.
- 03
Create the deployment
POST /v1/deployments with your image, GPU pool tier, and worker bounds. Or use the dashboard form — slug, image, tier, min/max workers, idle timeout. Live in under a minute.
- 04
Call POST /v1/gpu/run/{slug}
Send `{ "input": { ... } }` and receive an async `job_id` or a sync result. Stream status, fetch outputs as signed URLs, or have us POST them to your webhook.