Source: https://docs.sogni.ai/run-a-worker/fast-worker/running-on-akash/

# Run a Sogni Comfy Worker on Akash Network

Use a custom Stack Definition Language (SDL) file to request one NVIDIA GPU and run the Sogni worker on an Akash provider. Get your [Sogni worker key and Fast Worker NFT token ID](https://docs.sogni.ai/run-a-worker/fast-worker/running-sogni-fast-worker-remotely/#before-you-rent-a-gpu) first.

## [#](https://docs.sogni.ai/run-a-worker/fast-worker/running-on-akash/#prepare-the-deployment)Prepare the deployment

This example requests one RTX 4090, 64 GiB of system RAM, and separate persistent volumes for worker state and models. Change the GPU, RAM, and disk sizes to fit [the jobs you intend to serve](https://docs.sogni.ai/run-a-worker/fast-worker/hardware-requirements/).

The example pins the Docker image recommended in November 2026. Before deploying, replace the image reference with the tag returned by [Sogni’s recommended image endpoint](https://api.sogni.ai/v1/worker-images/comfy?cuda=13). The example bid ceiling under `pricing` is in `uact` per block. Set your own ceiling and review the rate the console shows before accepting a bid. [Akash pricing denomination](https://akash.network/docs/developers/deployment/akt/deployments/#author-an-sdl)

```
version: "2.0"

services:
  sogni-worker:
    image: sogni/comfy-worker-cu13:1.0.263
    env:
      - API_KEY=REPLACE_WITH_YOUR_SOGNI_WORKER_KEY
      - NFT_TOKEN_ID=REPLACE_WITH_YOUR_FAST_WORKER_NFT_ID
      - HEALTH_PROBE_PORT=8001
    expose:
      - port: 8001
        as: 80
        to:
          - global: true
    params:
      storage:
        worker-data:
          mount: /data
        worker-models:
          mount: /data-models

profiles:
  compute:
    sogni-worker:
      resources:
        cpu:
          units: 8
        memory:
          size: 64Gi
        storage:
          - size: 50Gi
          - name: worker-data
            size: 20Gi
            attributes:
              persistent: true
              class: beta3
          - name: worker-models
            size: 300Gi
            attributes:
              persistent: true
              class: beta3
        gpu:
          units: 1
          attributes:
            vendor:
              nvidia:
                - model: rtx4090
  placement:
    akash:
      pricing:
        sogni-worker:
          denom: uact
          amount: 1000

deployment:
  sogni-worker:
    akash:
      profile: sogni-worker
      count: 1
```

Akash requires a public `expose` entry. The example publishes only the worker’s dedicated health server on port `8001`. It provides health status without the control API. Leave the worker control port `8000` and ComfyUI port `8188` private. See the [Akash SDL reference](https://akash.network/docs/developers/deployment/akash-sdl/syntax-reference/) for GPU model names, volume classes, and pricing.

## [#](https://docs.sogni.ai/run-a-worker/fast-worker/running-on-akash/#select-a-provider-and-start)Select a provider and start

1.  In [Akash Console](https://console.akash.network/), start a new deployment and paste the SDL into the editor. Replace the placeholders only in your private deployment. The worker key is passed to the provider in the deployment manifest, so use a dedicated key and keep the configured SDL out of public repositories and messages. See [Akash environment variables and secrets](https://akash.network/docs/learn/core-concepts/environment-secrets/).
2.  Review the bids. Confirm the host supports **NVIDIA driver R580 or newer** and your requested storage class before accepting a bid. If you get no bids, check GPU availability, storage class, and your bid ceiling.
3.  Create the lease and open the deployment’s **Logs** tab. Watch for the worker to connect and download its models, then [verify the NFT in Sogni](https://docs.sogni.ai/run-a-worker/fast-worker/running-sogni-fast-worker-remotely/#verify-the-deployment).

The volumes can survive a container restart on the same lease. Back up anything you want to keep before closing the deployment or moving to another provider. When you finish, wait for the active job to complete, then use **Close Deployment** in Akash Console to end the lease. See [Akash’s deployment walkthrough](https://akash.network/docs/getting-started/quick-start/).
