Skip to main content
Page is under construction.

Check the github issues for ways to contribute! Or provide your feedback in this quick form
This guide details how to integrate your own AI models or GPU infrastructure into the Livepeer AI inference network. BYOC lets you run inference tasks with custom models (e.g. ControlNet, Whisper, SegFormer), deploy nodes across cloud, edge, or on-prem environments, and serve Livepeer inference workloads with economic incentives.
For full BYOC architecture, Gateway/Orchestrator roles, container lifecycle, and Protocol vs AI BYOC concepts, see the BYOC full guide.

​
Requirements

  • Linux or Docker-capable machine with GPU (CUDA 11+)
  • Internet-accessible IP or NAT hole-punching
  • Git, Python 3.9+; optional ComfyUI fork for modular tasks
  • Livepeer Gateway credentials or API key for worker registration

​
Step 1 — Clone and setup

git clone https://github.com/livepeer/comfystream
cd comfystream
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
Install your desired inference model(s): 
python scripts/download.py --model whisper-large
python scripts/download.py --model sdxl

​
Step 2 — Configure node

Edit your config.yaml: 
publicKey: "0xYourEthereumAddress"
gatewayURL: "wss://gateway.livepeer.org"
models:
  - whisper-large
  - sdxl
heartbeat: true
Optional: run as Docker; add plugin adapters (e.g. segment-anything, blip2).

​
Step 3 — Start gateway node

python run.py --adapter grpc --gpu --model whisper-large
You should see logs for heartbeats to the Livepeer Gateway, job claims and model execution, and result uploads or webhooks.

​
Step 4 — Register (optional)

Register your node on-chain (Arbitrum) so Gateways can discover and route work to you: 
livepeer-cli gateway register \
  --addr=1.2.3.4:5040 \
  --models=whisper-large,sdxl \
  --bond=100LPT \
  --region=NA1
Contract and ABI references are in the Blockchain contracts docs under About / Resources.

​
Pipeline examples

Whisper + caption overlay: 
pipeline:
  - task: whisper-transcribe
  - task: overlay
    type: caption
Blur faces + YOLO: 
pipeline:
  - task: object-detection
    model: yolov8
  - task: segment-blur
    target: faces

​
Troubleshooting

IssueFix

​
Developer notes

  • BYOC workers use the same Gateway Protocol as the rest of the Livepeer network. See Network technical architecture under About.
  • You can serve from multiple geographic regions.
  • Contribution rewards may be offered in LPT or credits; check the Forum and Builder opportunities.

​
See also

​
Resources

Last modified on February 18, 2026