Runflow
ComfyUI → Production API

Deploy your ComfyUI workflow as an API, in one-click.Unlock what's beyond it.

Deploy your ComfyUI workflow in seconds at half the market price. Launch production-ready pipelines through automatic quality checks, loops, and capabilities that ComfyUI alone can't give you.

Limited beta spots · No credit card required

See how it works ↓
One-click deploy from inside ComfyUI· Any custom node, any model· Auto-scaling GPU· Add loops, retries, and Sentinel (optional)
1-click
From ComfyUI to live API endpoint
$2.41
Per hour on RTX 4090 — billed by the second
17
Production-validated pipelines ready today
8
Quality dimensions Sentinel checks every generation

How it works

Deploy. Configure. Go live.
Call your API.

The same flow as any ComfyUI deployment platform — plus the layer of production capability they don't have.

01
One click from ComfyUI

Deploy your workflow directly from ComfyUI

From within ComfyUI, deploy your workflow directly to Runflow with a single click. We parse your node graph, resolve all custom node dependencies and models, and map your inputs and outputs automatically. Any workflow that runs in ComfyUI works here — custom nodes, LoRAs, multi-model chains, all of it.

→ Click "Deploy to Runflow" from within ComfyUI
✓ Parsed 24 nodes — FLUX.1-dev, ControlNet, IPAdapter
✓ Custom nodes resolved: 6 detected, all supported
✓ Inputs: image_url, style, background — mapped
02
Configure your pipeline

Add loops, retries, and Sentinel

This is where Runflow goes beyond standard deployment. Enable Sentinel quality evaluation, configure retry loops for failed outputs, and set your quality thresholds. No code required — all configured through the Runflow dashboard before you deploy.

Pipeline config — headshot-workflow
✓ Sentinel: enabled — 8 quality dimensions
✓ Auto-retry: on quality fail, max 3 attempts
✓ Loop: disabled (enable for batch workflows)
✓ GPU: RTX 4090 · $2.41/hr · billed per second
03
One click

Deploy — your endpoint is live with a click

Click deploy. Runflow provisions GPU, loads your models, and spins up your endpoint. You get a clean REST API with typed inputs matching your workflow parameters. Call it from any language or stack. Scale to zero when idle, burst to hundreds of parallel jobs on demand.

→ Deploy
✓ Models loaded into GPU memory
✓ Endpoint live: api.runflow.io/v1/run/headshot-workflow
✓ API docs auto-generated · ready to call
Endpoint live — one click from configuration

Ready to deploy your workflow?

Deploy your workflow with a click and get a live API with Sentinel and auto-retries — no configuration required.

Only on Runflow

Everything a deployment platform gives you.
Plus three things none of them have.

Deploying works everywhere. Loops, auto-retries, and Sentinel are only here.

🔄

Loops

Run your workflow in a loop, iterating until the output meets your quality criteria. Configure loop conditions directly in your pipeline — no code, no custom infrastructure.

Not available anywhere else
↩️

Auto-retries

If a generation fails or scores below threshold, Runflow automatically retries with the same or adjusted parameters. Set once in the dashboard. Never write retry logic again.

Zero code required
🛡

Sentinel quality evaluation

Every output scored across 8 quality dimensions before delivery. Identity, garment fit, background consistency, skin realism, and more — fully automated before the user sees anything.

Only platform outside enterprise GCP

Sentinel — quality evaluation

The only image quality layer
that understands intent.

Other platforms tell you when a generation fails technically. Sentinel tells you when it fails compared to the expected output, fully automated. Allowing you to fix it before your user sees it.

Stage 1

Intent understanding

Sentinel reads the workflow inputs and use case to build an evaluation plan. It understands what the output should achieve — not just what the prompt says.

Stage 2

Pre-processing — small models

Face similarity, multi-model segmentation, pose analysis, and other visual checks run first to prepare structured data for the judges. Each judge only receives what it needs.

Stage 3

Specialized LLM judges

One judge per quality dimension: identity, garment fit, background, logo fidelity (per logo instance), skin realism, text accuracy, model presence, zipper integrity. Each returns pass/fail with reasoning and confidence.

Sentinel run — virtual try-onPassed
INTENT: GARMENT REPLACEMENT, PRESERVE MODEL IDENTITY
STAGE 1 — PRE-PROCESSING
Face similarity✓ PASS0.97
Segmentation (garment)✓ PASS0.91
Pose analysis✓ PASS0.94
STAGE 2 — LLM JUDGES
Identity preservation✓ PASS0.96
Garment fit + drape✓ PASS0.88
Background consistency✓ PASS0.93
Skin realism⚠ RETRY0.71
Skin realism (retry 1)✓ PASS0.89
✓ DELIVERED — 4.1s · 1 auto-retry · 0 bad outputs to user

Platform features

Everything you expect from a
production ComfyUI platform.

The complete deployment layer — plus the three things that make Runflow different.

⬆️

One-click deploy from ComfyUI

Deploy your workflow to Runflow directly from within ComfyUI. Any custom node, any model, any LoRA. If it runs in ComfyUI, it runs here.

✓ Full custom node support

One-click deployment

Deploy your workflow and click deploy. You get a production REST endpoint with typed inputs matching your workflow parameters — no DevOps, no configuration, just a button.

✓ Auto-generated API docs
📈

Auto-scaling GPU — billed by the second

Workers scale to zero when idle. Burst to hundreds of parallel jobs instantly. You pay by the second of actual GPU runtime — at $2.41/hr on RTX 4090, with zero idle cost.

✓ Zero idle cost
🌍

RTX 4090 · A100 · L40S

Pick the GPU that fits your workflow — RTX 4090 for cost-efficient generation, A100 for high-throughput pipelines, L40S for the best price-to-performance ratio. Runflow auto-routes each job to your chosen hardware.

✓ Automatic hardware routing
📋

Version history

Every workflow update is versioned. Iterate with confidence, roll back to any previous version in one click, and maintain full history across your team.

✓ One-click rollback
🔁

Dev → Staging → Prod

Promote workflows across environments with the same controls developers expect. Test on staging before promoting to production — no surprises in live traffic.

✓ Full environment control
CASE STUDY

From 40% to 87% gross margin in 12 months

BetterPic processes hundreds of thousands of AI inference jobs monthly — portraits, background removal, segmentation, and more. Runflow's orchestration layer handles GPU routing, automatic failover, and retry across datacenters, so their team ships product instead of debugging infrastructure. This is the platform we're now opening to everyone.

30%+
savings vs in-house
87%
current gross margin
100K+
AI jobs per month
See Case Study

Margin Trend Analysis

Strong Upward Trend
0%30%60%90%Jan '25Mar '25May '25Jul '25Sep '25Nov '25Jan '26

Runflow vs other deployment platforms

Create production-ready workflows at scale from your local environment.

Runflow is the only platform that allows you to create, deploy and optimize your AI workflows at scale.

Other ComfyUI deployment platforms

Deploy and call API

The baseline — fast to deploy, nothing after

What they include

One-click deploy Deploy your workflow with a button, API live instantly.
Custom nodes and models Full environment support, any model stack.
Auto-scaling GPU Scales to zero, bursts on demand. At market rate pricing.

What's missing

No local deploy Requires uploading your workflow file manually. Can't deploy directly from inside ComfyUI.
No loops Iterative generation logic lives outside the platform, written by you.
No auto-retries Failed outputs ship. Your application handles recovery.
No quality evaluation No awareness of what outputs look like. Bad results reach users silently.

Runflow

Deploy, optimize, scale

Everything above — plus the production quality layer

Included as standard

Deploy from local ComfyUI Click deploy from inside your local ComfyUI instance — no file upload, no export. API live instantly.
Custom nodes and models Any workflow that runs in ComfyUI runs here.
$2.41/hr on RTX 4090 Billed by the second, zero idle cost.

Only on Runflow

Loops built in Configure iterative generation directly in your pipeline. No code.
Auto-retries on failure or quality threshold Set once in the dashboard, never write retry logic again.
Sentinel quality evaluation Every output scored across multiple dimensions before delivery. Only platform with this capability.

Supported workflows

Start from a validated pipeline.
Or bring your own.

Every pre-built pipeline runs with Sentinel and auto-retries enabled by default. Deploy any workflow and the same production layer applies.

👤
AI Headshots
People Imagery
Live
👗
Virtual Try-On
On-Model / Fashion
Live
📣
Ad Creative Generator
Ad Creative
Live
📦
Product Photography
Product Imagery
Live
✂️
Advanced Segmentation
Utility
Live
Bring your own
Any ComfyUI workflow
Deploy →

View all 17 pipelines →

The pro ComfyUI platform

Ship your workflow.
Unlock what's beyond it.

Deploy your ComfyUI workflow with a click of a button at half the market price. Add loops, quality checks, and capabilities that ComfyUI alone can't give you.

Limited beta spots · No credit card required · $2.41/hr RTX 4090 · Loops + Sentinel included