BlockFlow

Open-source visual pipelines for AI image and video generation.

BlockFlow's main path runs ComfyUI workflows through ComfyGen on your own RunPod serverless endpoint. Provider-specific blocks let you mix in models from RunPod, PiAPI, OpenRouter, CivitAI, and more.

npx @hearmeman24/blockflow

Pipeline Editor

ComfyUI Gen Block

Artifacts

Why BlockFlow

ComfyUI is powerful, but production workflows often need more than one graph run. You may need prompt generation, reference images, a ComfyUI workflow, post-processing, review, publishing, and repeated runs across many inputs.

BlockFlow turns that into a left-to-right pipeline. Blocks can branch, but you do not have to wire low-level graph nodes together for every production flow.

The scale model is RunPod serverless: if your endpoint has 10 available workers, BlockFlow can submit work in parallel instead of waiting for one local GPU queue. You can also run multiple pipeline tabs at the same time, each with its own state, cancellation, outputs, and artifacts.

Generation Backends

BlockFlow can orchestrate multiple generation backends in one pipeline.

ComfyUI via ComfyGen

The primary backend path is ComfyUI through ComfyGen. BlockFlow provisions or attaches a RunPod serverless endpoint, sends ComfyUI workflows to it, monitors jobs, handles cancellation, and stores outputs locally.

This is the path for:

• arbitrary ComfyUI workflow JSONs
• installed workflow/model presets
• LoRA-aware ComfyUI generation
• model downloads to the endpoint's network volume
• serverless worker scaling

Direct Provider Blocks

Some blocks call hosted models directly instead of going through ComfyUI:

• Nano Banana 2 on RunPod
• Seedance 2 through PiAPI
• GPT Image through PiAPI
• Prompt and multimodal prompt writing through OpenRouter
• CivitAI sharing for publishing generated media

You can mix these in one pipeline: generate a prompt, create or edit an image with one provider, animate it with another, upscale it, review it, and publish the result.

Presets

BlockFlow can install ComfyGen presets: packaged workflow + model bundles that land on your own ComfyUI RunPod endpoint.

The public preset registry lives here:

github.com/Hearmeman24/blockflow-presets

Current examples include:

• gbrx-mop-pro
• hidream-o1
• ltx-2-3
• qwen-image-lighting
• wan-animate
• wan22-svi-4pass

Presets are useful when a workflow needs a specific model set, hidden internal nodes, exposed user controls, or repeatable setup across machines.

What You Can Build

• ComfyUI generation pipelines backed by RunPod serverless workers
• prompt -> image -> video -> upscale flows
• image and video reference workflows
• dataset creation and captioning flows
• LoRA training and upload-to-ComfyGen flows
• batch and sweep-style runs across prompts, LoRAs, settings, or references
• human review gates before downstream steps
• artifacts that can be restored, inspected, or submitted to CivitAI

Quick Start

Run the published package:

npx @hearmeman24/blockflow

BlockFlow starts a local FastAPI backend and a prebuilt Next.js frontend, then opens the browser UI.

On first use:

1. Open Settings and add the credentials for the services you want to use.
2. Set up or attach a ComfyGen RunPod endpoint.
3. Install a preset, paste a ComfyUI workflow, or add direct provider blocks.
4. Build a pipeline and click Run Pipeline.

BlockFlow runs locally and uses your own API keys, RunPod account, and storage. Generation costs are paid directly to the services you connect.

Common Workflows

ComfyUI at Serverless Scale

Install a preset or paste a workflow, configure the exposed controls, then run it through a ComfyGen endpoint. Increase the endpoint worker count when you need more parallel generation.

Content Pipelines

Use prompt blocks, image/video generation blocks, viewers, upscalers, and publishing blocks as one repeatable workflow instead of a folder of disconnected scripts.

LoRA and Dataset Workflows

Create datasets, caption images, submit LoRA training jobs, and upload trained LoRAs back to the ComfyGen endpoint so downstream generation can use them.

Local Development

For repository development:

git clone https://github.com/Hearmeman24/BlockFlow.git
cd BlockFlow
uv run app.py

The dev entrypoint starts FastAPI on :8000 and Next.js on :3000.

Useful commands:

uv run pytest
npm --prefix frontend test
npm --prefix frontend run build

Documentation

• Architecture
• Testing
• ComfyGen sidecar resolution
• npm release flow
• Private blocks
• Contributing

License

MIT. See LICENSE.