​
On-chain Setup

In the previous section, we covered the off-chain configuration of your AI Gateway node. This section will explain how to connect your AI Gateway to the Livepeer AI network to start requesting AI inference jobs on-chain.

​
Prerequisites

  • A dedicated static IP address or domain name for your AI Gateway node
  • A funded Ethereum account with enough ETH to cover gas fees and AI inference job payments

​
Launching your On-chain AI Gateway

Once your AI Gateway is set up off-chain, it’s time to connect it to the Livepeer AI network for on-chain inference jobs. The process is similar to setting up a Mainnet Transcoding Network Gateway, with a few extra flags specific to Livepeer AI. You can refer to the Mainnet Transcoding Network Gateway guide for more information. Here’s a summary of the key steps:

1

Verify Off-chain AI Gateway

Ensure your AI Gateway node is functioning correctly off-chain before connecting it to the Livepeer AI network. Refer to the previous section for details.

2

Prepare your Ethereum account

Make sure your Ethereum account has enough ETH to cover gas fees and AI inference job payments. For more details, see the Fund Gateway Guide.

3

Fund your AI Gateway

Deposit enough funds into your AI Gateway to cover the deposit and reserve needed for AI inference requests. Follow the steps in the Deposit Gateway Funds via Livepeer CLI guide.

4

Configure Transcoding Options

Configure transcoding options even though the AI Gateway doesn’t handle transcoding, due to current Livepeer AI software requirements. See the Transcoding Options guide for more information.

5

Launch the AI Gateway

Now launch your AI Gateway node. The process is similar to the docker or binary setup in the previous section, but requires additional flags to enable on-chain functionality. Example:

  • -aiServiceRegistry: Ensures that the Gateway is connected to the Livepeer AI network.
  • -network=arbitrum-one-mainnet: Connects the AI Gateway to the Arbitrum Mainnet network.
  • -ethUrl=https://arb1.arbitrum.io/rpc: Sets the Arbitrum Mainnet RPC URL. Replace it with your own if necessary.
  • -ethKeystorePath=/root/.lpData/arbitrum-one-mainnet/keystore: Sets the path to the Ethereum keystore file for the AI Gateway.
  • -ethAcctAddr <LIVEPEER_AI_ORCH_ETH_ADDRESS>: Sets the Ethereum account address for the AI Gateway.
  • -ethPassword=/root/.lpData/.eth_secret: Sets the path to the Ethereum keystore password file.
  • -ethOrchAddr=<MAIN_ORCH_ETH_ADDRESS>: Sets the Ethereum address of the Mainnet Transcoding Network Gateway.
  • -maxTotalEV=100000000000000: Ensures the AI Gateway complies with the max ticket value limits in go-livepeer.
  • -maxPricePerUnit=<MAX PRICE in WEI or USD>: This defaults to 0. Set to acceptable max price in wei (or USD, e.g. 0.02USD) willing to pay. Note: this can impact amount of Orchestrators available to process the work.
  • -ignoreMaxPriceIfNeeded=<true or false>: This defaults to false. Set to true if want to process requests if no Orchestrators are under maxPricePerUnit or do not want to use a max price.
  • -maxPricePerCapability=/path/to/maxPrices.json: This flag sets the max price per unit for one or many pipeline/models. Refer to the Set Max Price Per Pipeline and Model section for more details.

​
Set Max Price Per Pipeline and Model (Optional)

To avoid overpaying for AI inference jobs, you can set a maximum price per unit for each pipeline and model. This feature is optional and can be configured using the -maxPricePerCapability flag. If a price is not set for a specific pipeline/model, the -maxPricePerUnit price will be used. The flag input should be a JSON file with the specified max prices. The example below demonstrates the following configurations:

  • Set the price for the image-to-image pipeline and model ByteDance/SDXL-Lightning to 1,700,000 wei per unit (pixels).
  • Set the price for the text-to-image pipeline and model stabilityai/stable-diffusion-3-medium-diffusers to 4,768,371 wei per unit (pixels).
  • Set the price for the upscale pipeline for all models to 4,768,371 wei per unit. The pixels_per_unit defaults to 1 if not specified.
  • Set the price for the image-to-video pipeline for all models to 3,390,842 wei per unit (pixels).
  • Set the price for the audio-to-text pipeline for all models to 12,882,811 wei per unit (milliseconds of audio).
{
  "capabilities_prices": [
    {
      "pipeline": "image-to-image",
      "model_id": "ByteDance/SDXL-Lightning",
      "price_per_unit": 1700000,
      "pixels_per_unit": 1
    },
    {
      "pipeline": "text-to-image",
      "model_id": "stabilityai/stable-diffusion-3-medium-diffusers",
      "price_per_unit": 4768371,
      "pixels_per_unit": 1
    },
    {
      "pipeline": "upscale",
      "model_id": "default",
      "price_per_unit": 4768371
    },
    {
      "pipeline": "image-to-video",
      "price_per_unit": 3390842,
      "pixels_per_unit": 1
    },
    {
      "pipeline": "audio-to-text",
      "price_per_unit": 12882811
    }
  ]
}

​
On-chain Setup

In the previous section, we covered the off-chain configuration of your AI Gateway node. This section will explain how to connect your AI Gateway to the Livepeer AI network to start requesting AI inference jobs on-chain.

​
Prerequisites

  • A dedicated static IP address or domain name for your AI Gateway node
  • A funded Ethereum account with enough ETH to cover gas fees and AI inference job payments

​
Launching your On-chain AI Gateway

Once your AI Gateway is set up off-chain, it’s time to connect it to the Livepeer AI network for on-chain inference jobs. The process is similar to setting up a Mainnet Transcoding Network Gateway, with a few extra flags specific to Livepeer AI. You can refer to the Mainnet Transcoding Network Gateway guide for more information. Here’s a summary of the key steps:

1

Verify Off-chain AI Gateway

Ensure your AI Gateway node is functioning correctly off-chain before connecting it to the Livepeer AI network. Refer to the previous section for details.

2

Prepare your Ethereum account

Make sure your Ethereum account has enough ETH to cover gas fees and AI inference job payments. For more details, see the Fund Gateway Guide.

3

Fund your AI Gateway

Deposit enough funds into your AI Gateway to cover the deposit and reserve needed for AI inference requests. Follow the steps in the Deposit Gateway Funds via Livepeer CLI guide.

4

Configure Transcoding Options

Configure transcoding options even though the AI Gateway doesn’t handle transcoding, due to current Livepeer AI software requirements. See the Transcoding Options guide for more information.

5

Launch the AI Gateway

Now launch your AI Gateway node. The process is similar to the docker or binary setup in the previous section, but requires additional flags to enable on-chain functionality. Example:

  • -aiServiceRegistry: Ensures that the Gateway is connected to the Livepeer AI network.
  • -network=arbitrum-one-mainnet: Connects the AI Gateway to the Arbitrum Mainnet network.
  • -ethUrl=https://arb1.arbitrum.io/rpc: Sets the Arbitrum Mainnet RPC URL. Replace it with your own if necessary.
  • -ethKeystorePath=/root/.lpData/arbitrum-one-mainnet/keystore: Sets the path to the Ethereum keystore file for the AI Gateway.
  • -ethAcctAddr <LIVEPEER_AI_ORCH_ETH_ADDRESS>: Sets the Ethereum account address for the AI Gateway.
  • -ethPassword=/root/.lpData/.eth_secret: Sets the path to the Ethereum keystore password file.
  • -ethOrchAddr=<MAIN_ORCH_ETH_ADDRESS>: Sets the Ethereum address of the Mainnet Transcoding Network Gateway.
  • -maxTotalEV=100000000000000: Ensures the AI Gateway complies with the max ticket value limits in go-livepeer.
  • -maxPricePerUnit=<MAX PRICE in WEI or USD>: This defaults to 0. Set to acceptable max price in wei (or USD, e.g. 0.02USD) willing to pay. Note: this can impact amount of Orchestrators available to process the work.
  • -ignoreMaxPriceIfNeeded=<true or false>: This defaults to false. Set to true if want to process requests if no Orchestrators are under maxPricePerUnit or do not want to use a max price.
  • -maxPricePerCapability=/path/to/maxPrices.json: This flag sets the max price per unit for one or many pipeline/models. Refer to the Set Max Price Per Pipeline and Model section for more details.

​
Set Max Price Per Pipeline and Model (Optional)

To avoid overpaying for AI inference jobs, you can set a maximum price per unit for each pipeline and model. This feature is optional and can be configured using the -maxPricePerCapability flag. If a price is not set for a specific pipeline/model, the -maxPricePerUnit price will be used. The flag input should be a JSON file with the specified max prices. The example below demonstrates the following configurations:

  • Set the price for the image-to-image pipeline and model ByteDance/SDXL-Lightning to 1,700,000 wei per unit (pixels).
  • Set the price for the text-to-image pipeline and model stabilityai/stable-diffusion-3-medium-diffusers to 4,768,371 wei per unit (pixels).
  • Set the price for the upscale pipeline for all models to 4,768,371 wei per unit. The pixels_per_unit defaults to 1 if not specified.
  • Set the price for the image-to-video pipeline for all models to 3,390,842 wei per unit (pixels).
  • Set the price for the audio-to-text pipeline for all models to 12,882,811 wei per unit (milliseconds of audio).
{
  "capabilities_prices": [
    {
      "pipeline": "image-to-image",
      "model_id": "ByteDance/SDXL-Lightning",
      "price_per_unit": 1700000,
      "pixels_per_unit": 1
    },
    {
      "pipeline": "text-to-image",
      "model_id": "stabilityai/stable-diffusion-3-medium-diffusers",
      "price_per_unit": 4768371,
      "pixels_per_unit": 1
    },
    {
      "pipeline": "upscale",
      "model_id": "default",
      "price_per_unit": 4768371
    },
    {
      "pipeline": "image-to-video",
      "price_per_unit": 3390842,
      "pixels_per_unit": 1
    },
    {
      "pipeline": "audio-to-text",
      "price_per_unit": 12882811
    }
  ]
}