Sogni: Learn logo
Running Sogni Fast Worker Remotely cover image

☁️Running Sogni Fast Worker Remotely

Don't have your own local Nvidia box to run Sogni Fast Worker with? This Worker is also designed to be deployable to any cloud GPU rental service which supports loading custom Docker images.

Salad.com hosts a Sogni Fast Worker Recipe. The older Sogni Fast Worker for Flux recipe is retired and replaced with the Sogni Comfy Worker. Choose Comfy Worker for Flux-family workflows, or Stable Diffusion Worker for Stable Diffusion models.

Sogni Fast Worker supports running on Distributed GPU Cloud Platforms. Nosana.com and Salad.com integrations have been well tested. Your mileage may vary with other platforms.

#Running on Salad.com

  1. Create a Sogni account if you don't have one already. You may create one at app.sogni.ai. Multiple Sogni Fast Workers can be run from one account or separate accounts.
  2. Log in to nft.sogni.ai and mint a free Sogni NFT for each GPU worker you want to deploy. Record the NFT Token ID and the API Key for each worker. The API Key is unique to the Sogni Account. The NFT Token ID will be unique for each worker.
  3. Sign up for a Salad account using the "Deploy on SaladCloud" option and follow their website instructions to create a new Container Group.
  4. Select "Deploy a Container Group" under the SaladCloud Organization screen.
  5. Select the "Sogni Fast Worker" Recipe under the Recipes section.
  6. Select your Sogni Worker Type, and add your Sogni API Key and NFT Token ID.
  7. Leave "Replicas" as 1.
  8. All other settings have been automatically configured for you. Simply press "Deploy" and wait for your remote Sogni Fast Worker deployment to come online!

#Running on other hosted GPU platforms

#Choose the correct Comfy Worker image

Use sogni/comfy-worker-cu13:latest for maximum performance when both of these conditions are true:

  • The GPU is an NVIDIA GeForce RTX 5090, NVIDIA RTX PRO 6000 Blackwell, another compatible Blackwell GPU, or a GeForce RTX 4090-series GPU.
  • The host uses NVIDIA driver branch R580 or newer.

The CUDA 13 image is typically 10–20% faster on RTX 4090-series GPUs and 30–50% or more faster on Blackwell GPUs, depending on the workflow. Use the compatibility image, sogni/comfy-worker:latest, for other GPUs or older drivers. The CUDA Version shown by nvidia-smi is the newest CUDA runtime the driver can support; verify that the installed driver itself is R580 or newer.

The official installer detects eligible hardware and switches between these two managed latest images automatically. Custom images and version-pinned tags are not changed. On a mixed-GPU host, it uses the CUDA 13 image only when every visible GPU is compatible.

Platforms that require an immutable, versioned image can query these endpoints:

Each endpoint returns the current versioned Docker image path for that runtime.

  1. Verify that your target platform supports running Docker container images with GPU access with a supported Nvidia GPU. It must also support configuring ENV Vars for the deployed instance.
  2. Create a Sogni account if you don't have one already. You may create one at app.sogni.ai. Multiple Sogni Fast Workers can be run from one account or separate accounts.
  3. Log in to nft.sogni.ai and mint a free Sogni NFT for each GPU worker you want to deploy. Record the NFT Token ID and the API Key for each worker. The API Key is unique to the Sogni Account. The NFT Token ID will be unique for each worker.
  4. For "Container Configuration," set the Image Source to sogni/comfy-worker-cu13:latest on the eligible R580+ Ada/Blackwell systems listed above. Use sogni/comfy-worker:latest on other systems. The old sogni/sogni-flux-worker:latest image is retired and replaced with Comfy Worker. Use sogni/sogni-stable-diffusion-worker:latest for a Stable Diffusion worker.
  5. Under Replica Count you will always leave this at 1. If you are running multiple workers you will need to create a new Container Group for each worker as Salad.com passes Environment Variables at the Container Group level.
  6. For vCPUs select 4. For Memory select 38 GB. For GPU select "RTX 5090". For Disk Space, select the maximum size available for model caching. As of July 2026, Sogni Comfy Worker needs about 470 GB to automatically download all supported Comfy models on RTX 5090 and higher-VRAM GPUs (32GB+ VRAM), or about 435 GB on 24GB GPUs like RTX 4090. Add extra space for Docker images, temporary downloads, logs, and the operating system. If there is a shm-size option, set it to 8g. The RTX 5090's 32 GB of VRAM meets the minimum for MiniMax H3 video and uses Sogni's Blackwell-specific runtime tuning. Other supported NVIDIA GPUs with at least 32 GB of VRAM may also host H3.
  7. For Health Check Probes: Startup Probe, select Protocol: "HTTP/1.X" and Path "/startup". Port 8000. Initial Delay Seconds: 90. Period Seconds: 5, Timeout Seconds: 5, Success Threshold: 2, Failure Threshold: 120.
  8. For Liveness Probe select Protocol: "HTTP/1.X" and Path "/liveness". Port 8000. Initial Delay Seconds: 90. Period Seconds: 10, Timeout Seconds: 30, Success Threshold: 1, Failure Threshold: 6.

    Note: If your platform does not support health probes you can skip steps 7 and 8; Sogni Worker has its own internal health probes as a fallback.
  9. Under "Environment Variables" add the following:
    • API_KEY: The API Key for your Sogni Account.
    • NFT_TOKEN_ID: The NFT Token ID for your Sogni worker.
    • AUTO_DOWNLOAD_TO_MIN_MODEL_COUNT: The number of models to download when the Stable Diffusion worker starts. The retired Flux worker has been replaced with Comfy Worker, which uses workflow-based configuration instead.
    • DISABLE_AUTOMATIC_DOWNLOADS=1 (optional, Comfy Worker): Serve only workflows and user LoRAs already present on the mounted model volume.
    • DISABLE_LORA_DOWNLOADS=1 (optional, Comfy Worker): Keep workflow model downloads enabled but serve only user LoRAs already present on the mounted model volume.
    • DATA_DOG_API_KEY: You may supply an optional DataDog API Key for troubleshooting. See Advanced Configuration for precedence, storage-limit, and multi-GPU details.
  10. Click "Deploy" and your worker will start. You check for latest integrations documentation at docs.sogni.ai and reach out at [email protected] for any additional assistance.

#Need help? Join our Discord!

If you have any issues or questions setting up your Sogni Fast Worker you can reach out on Discord or via email to [email protected] for technical support.

Join Sogni Discord ✨

Last updated 2026-08-10