Sogni Fast Worker Recipe

Last Updated: July 13, 2026

​

Overview

The Sogni Fast Worker recipe deploys a GPU worker that connects to the Sogni decentralized inference network. The worker processes image or video inference jobs and may earn rewards for successfully completed work under Sogni’s current eligibility requirements and terms. The recipe offers two worker types:

• Comfy Worker - Optimized for inference with the latest open-weight image and video models on the Sogni network. It uses sogni/comfy-worker:latest and is the default option.
• Stable Diffusion Worker - Supports Stable Diffusion community models and ControlNet workflows. It uses sogni/sogni-stable-diffusion-worker:latest.

See About Sogni Fast Worker for the current model support, GPU requirements, and worker policies.

​

Prerequisites

Before deploying the recipe, you need:

1. A Sogni account.
2. A Sogni API key.
3. A unique Sogni Fast Worker NFT Token ID for this GPU worker.
4. A SaladCloud account with access to the configured GPU class.

Follow Sogni’s remote worker guide to create the required Sogni credentials.

​

Quick Start

1. Open the SaladCloud Portal.
2. Deploy the Sogni Fast Worker recipe.
3. Enter a Container Group Name.
4. Choose Comfy or Stable Diffusion as the worker type.
5. Enter your Sogni API Key and Sogni NFT Token ID.
6. Keep the replica count at 1, then deploy the container group.
7. Monitor the container logs until the worker authenticates and connects to the Sogni network.

​

Recipe Defaults

The recipe preserves the established Sogni worker configuration:

• Replicas: 1
• vCPUs: 8
• Memory: 38 GB
• Storage: 250 GB
• Shared memory: 2 GB
• Priority: high
• Image caching: enabled
• Startup probe: GET /startup on port 8000
• Liveness probe: GET /liveness on port 8000

The recipe maps the form credentials to these container environment variables:

• API_KEY - your Sogni account API key
• NFT_TOKEN_ID - the unique NFT Token ID assigned to this worker

The Stable Diffusion option also sets AUTO_DOWNLOAD_TO_MIN_MODEL_COUNT from the Models to Download at Start field. Comfy Worker downloads compatible workflow assets as jobs require them and does not use that Stable Diffusion setting.

​

Replica Count And Credentials

Use exactly one replica per deployment. SaladCloud sets environment variables at the container group level, while each Sogni GPU worker requires its own NFT Token ID. To run multiple workers, create a separate container group with a unique NFT Token ID for each worker. The worker connects outbound to Sogni and does not expose a public inference API through the Salad Container Gateway. Authentication is handled by the Sogni API key and NFT Token ID.

​

Sogni Unlimited

Sogni is rolling out its Sogni Unlimited fair-use subscription program. Eligible Fast Workers can participate in subscription-covered jobs after the associated Sogni account accepts the current terms and enables participation. Eligibility, job allocation, and revenue sharing are controlled by Sogni and may change independently of this SaladCloud recipe. Review the Sogni Fast Worker FAQ for current program requirements and worker revenue information.

​

Startup And Monitoring

The container image and model or workflow assets can take tens of minutes to download, depending on the selected worker, node bandwidth, and which assets are already cached. The startup probe allows up to approximately ten minutes after its initial delay before SaladCloud considers startup unsuccessful. After deployment:

1. Wait for the instance to show a green checkmark in the Ready column.
2. Inspect container logs for successful Sogni authentication and network connection messages.
3. Use Sogni’s worker dashboard and analytics to confirm that the NFT Token ID appears online and receives jobs.

Sogni publishes worker updates under the latest image tags. If Sogni reports that a worker version is no longer supported, edit the container group and resubmit the image path to force the platform to refresh the image, then restart the worker.

​

Source Code

• Recipe Source
• Sogni Fast Worker documentation
• Running Sogni Fast Worker remotely
• Sogni Fast Worker FAQ