Quickstart: ComfyStream on Livepeer

By the end of this quickstart, you will have a ComfyStream instance running with a real-time AI effect applied to a live video feed. Once you have a working pipeline, the final section shows how to connect it to the Livepeer network.

Before You Start

ComfyStream processes live video using ComfyUI workflows on a local or cloud GPU. Choose your deployment path based on what you have available:

Path	Best for	Requires
RunPod	Fastest start, no local GPU	RunPod account (~$0.50/hr for A40)
Docker	Local GPU or cloud server	Docker + NVIDIA GPU, Linux
Local install	Existing ComfyUI setup	Miniconda, NVIDIA GPU, Linux

ComfyStream requires an NVIDIA GPU. Windows and macOS are not supported for the server component. //: # (REVIEW: Confirm Linux-only from docs.comfystream.org or Rick. The README does not explicitly state OS, but PyTorch + CUDA dependency strongly implies Linux/NVIDIA only.)

Prerequisites across all paths:

NVIDIA GPU with sufficient VRAM //: # (REVIEW: Verify minimum VRAM from docs.comfystream.org hardware section. Likely 12–16 GB for StreamDiffusion; 24 GB recommended for real-time performance.)
A modern browser (Chrome or Firefox) for the ComfyStream UI

Set Up ComfyStream

RunPod (fastest)
Docker (local or cloud server)
Local install

Deploy the RunPod template

Open the livepeer-comfystream RunPod template and select a GPU pod.For StreamDiffusion workflows, select a GPU with at least //: # (REVIEW: confirm VRAM) VRAM. An RTX A4000 or A40 is a reasonable starting point.Click Deploy.

Wait for the pod to start

Once the pod status shows Running, click Connect to open the pod’s exposed ports.ComfyStream exposes two ports:

8188 — ComfyUI interface
8889 — ComfyStream WebRTC server //: # (REVIEW: Confirm exact ports from docs.comfystream.org or docker-compose.yml in the repo. These are based on standard ComfyUI port + common ComfyStream server port.)

Open the ComfyStream UI

In the RunPod connect panel, open the HTTP service on port 8889. This loads the ComfyStream browser UI.You should see a camera input selector and a workflow panel.

Pull the Docker image

docker pull livepeer/comfystream

Run the container

docker run --gpus all \
  -p 8188:8188 \
  -p 8889:8889 \
  livepeer/comfystream

For full installation options including volume mounts and workspace configuration, see the ComfyStream install docs. //: # (REVIEW: Confirm exact docker run flags from docs.comfystream.org / docker-compose.yml. The above is a minimal illustration only — actual flags may differ (workspace path, model path mounts).)

Open ports for WebRTC (remote server only)

If running on a remote server, allow inbound and outbound UDP traffic on ports 1024–65535. This is required for WebRTC peer connections.If only a subset of ports is available, pass --media-ports 1024,1025,... to the server command.

Open the ComfyStream UI

Open http://localhost:8889 (local) or your server’s public IP on port 8889. The ComfyStream UI should load.

Create a Conda environment

conda create -n comfystream python=3.12
conda activate comfystream

Install ComfyStream

pip install git+https://github.com/livepeer/comfystream.git

Install custom nodes

If you are running ComfyStream as a standalone application (not inside ComfyUI), copy the auxiliary custom nodes into your ComfyUI custom_nodes folder. See the ComfyStream README for the exact list.

Download models

python src/comfystream/scripts/setup_models.py --workspace /path/to/ComfyUI

Start the server

python server/app.py --workspace /path/to/ComfyUI

The server starts on port 8889 by default. Open http://localhost:8889 in your browser.

Load a Workflow

ComfyStream uses ComfyUI workflow JSON files. The repository includes multiple starter workflows under the workflows/ directory.

Open the workflow panel

In the ComfyStream UI, click the workflow selector and choose a workflow file.For your first run, use the StreamDiffusion SD 1.5 workflow — it is the lightest and fastest to compile. //: # (REVIEW: Confirm the recommended starter workflow filename from the repo’s workflows/ directory. Likely something like streamdiffusion_sd15.json.)

Select your camera input

Choose your webcam from the camera input dropdown. The UI will request camera permission.

Start the pipeline

Click Run. The first run requires TensorRT compilation, which takes 2–5 minutes. Subsequent runs load immediately.You will see progress logs in the terminal where the server is running.

Verify the Pipeline is Running

Once compilation completes, the ComfyStream UI will show your webcam feed with the AI effect applied in near-real-time. Expected result: Your webcam input appears transformed by the workflow — style transfer, depth-mapped effects, or other visual processing depending on the workflow you loaded. If you see only the raw webcam feed without transformation, check:

GPU VRAM is not exhausted (check nvidia-smi)
The workflow compiled without error (check server logs)
The workflow nodes reference models that have been downloaded

Want to see what a running pipeline looks like before setting up? The Building Real-Time AI Video Effects with ComfyStream post includes a 5-minute demo video.

Connect to the Livepeer Network

Running ComfyStream locally gets you a working real-time AI pipeline. To deploy that pipeline on the Livepeer network — making it accessible to other applications or earning compute fees — there are two paths:

Path	What it does	When to use
Daydream API	Use Livepeer Inc’s hosted ComfyStream infrastructure	You want to serve your pipeline without managing compute
BYOC worker	Register your ComfyStream instance as a go-livepeer orchestrator	You want to earn fees and run production workloads on the network

For the Daydream API, request access at daydream.live. For the BYOC path, the integration layer is PyTrickle — a Python package that enables ComfyStream to register as a Livepeer AI worker. See the BYOC documentation for setup steps.

What You Can Build

ComfyStream supports the following pipeline types in production (Phase 4, January 2026):

StreamDiffusion — real-time style transfer and image-to-image on live video
StreamDiffusion V2 — second-generation diffusion pipeline, supports video-to-video and image-to-image
SuperResolution — real-time video upscaling
AudioTranscription + SRT — real-time captions embedded in video output
Text data-channel output — structured text output (e.g. transcription) alongside video

Next Steps

Build with ComfyStream

Custom nodes, workflow design, and advanced ComfyStream configuration.

Bring Your Own Container

Package any Python workload as a Livepeer AI worker using the BYOC path.

ComfyStream documentation

Full ComfyStream node reference, troubleshooting, and hardware guides.

Hub

Concepts

Get Started

Build

Guides

Opportunities

Resources

Quickstart: ComfyStream on Livepeer

Before You Start

Set Up ComfyStream

Load a Workflow

Verify the Pipeline is Running

Connect to the Livepeer Network

What You Can Build

Next Steps

Build with ComfyStream

Bring Your Own Container

ComfyStream documentation

Hub

Concepts

Get Started

Build

Guides

Opportunities

Resources

​Before You Start

​Set Up ComfyStream

​Load a Workflow

​Verify the Pipeline is Running

​Connect to the Livepeer Network

​What You Can Build

​Next Steps

Build with ComfyStream

Bring Your Own Container

ComfyStream documentation

Before You Start

Set Up ComfyStream

Load a Workflow

Verify the Pipeline is Running

Connect to the Livepeer Network

What You Can Build

Next Steps