Comfy Rail Book a technical fit review

Run your AI SaaS
on your own API.

Deploy your existing ComfyUI workflows as maintained RunPod runtimes — in your own account, behind your own API, with infrastructure costs under your control.

Book a technical fit review See ProductBackground in production

Independent commercial runtime for ComfyUI deployments on RunPod.
Not affiliated with or endorsed by Comfy Org or RunPod.

Generation

Generated Alpine landscape sample
workflow: gen_v1
runtime: gpu-24gb
contract: api-v1

Outpainting

Expanded Alpine landscape sample
workflow: outpaint_v1
runtime: gpu-48gb
contract: api-v1

Upscaling

Detailed Alpine landscape sample
workflow: upscale_v1
runtime: gpu-24gb
contract: api-v1
POST /v1/runYour API endpoint

Built for teams with production workflows

Comfy Rail runs your ComfyUI workflows as maintained runtimes on RunPod — in your account, on your terms.

Three runtimes. One stable contract.

Generation, outpainting and upscaling ship as separate, versioned images. Your workflow remains portable inside API Contract v1.

01

Generation

For initial image creation and smaller production workflows.

02

Outpainting

For controlled canvas expansion and larger image pipelines.

03

Upscaling

For 2× and 4× production output.

Know exactly what runs.

Released profiles share a pinned Z-Image Turbo and Qwen3 foundation, UltraFlux VAE, the first-party HBGv4-4250 LoRA, NMKD Superscale and the CreativeML OpenRAIL-M safety checker. Profile-specific roles appear below; exact revisions, file hashes, licenses and use restrictions are recorded in the release manifest.

01

Generation

Initial image creation and refinement

Image path

Z-Image Turbo
BF16 diffusion core
UltraFlux VAE
Image encode and decode
HBGv4-4250 LoRA
First-party generation refinement

Prompt and output

Qwen3-4B
Text conditioning
Qwen3-4B-Instruct-2507
Prompt transformation
4× NMKD Superscale
Final output enhancement
02

Outpainting

Controlled canvas expansion

Expansion path

Qwen-Image
Quantized diffusion core and VAE
InstantX ControlNet
Masked inpainting control
Qwen-Image Lightning
Eight-step acceleration adapter

Scene guidance

Qwen3-VL-4B
Visual prompt interpretation
DA3 Metric Large
Depth-aware structure guidance
Z-Image ControlNet Union 2.1
Secondary controlled refinement
03

Upscaling

2× and 4× production output

2× restoration path

FlashVSR v1.1
Diffusion-based restoration
DA3 Metric Large
Structure and depth estimation

4× refinement path

4× NMKD Superscale
Learned final enlargement
Z-Image Turbo + HBGv4
Controlled detail refinement
FlashVSR v1.1
First restoration stage

GPU qualification targetsRTX 4090 · RTX 5090 · L40S · RTX PRO 6000 Blackwell

Release controlsImmutable revisions · SHA256 verification · no model downloads at startup

License evidencePer-model provenance, license notices and applicable use restrictions

Commercial boundaryOnly release-gated models and nodes enter customer images

Your account. Your endpoints. Your data path.

01

Your backend

RunPod bills you directly.

02

RunPod API

Submit through POST /run.

03

Private runtime image

Images deploy by immutable digest.

04

Customer storage

Presigned uploads keep outputs in your storage.

Request payload (JSON)
{
  "workflow_json": { "nodes": { "...": "..." } },
  "input_file": "data:image/png;base64,...",
  "seed": 123456789,
  "upload_url": "https://storage.example/presigned-put",
  "output_format": "webp"
}

Built for teams with a
production workflow.

Fits if:

  • You already run one to three complex ComfyUI workflows.
  • A paid product, client project or launch date depends on them.
  • You want self-hosting on RunPod and control of models and customer data.
  • You have a technical owner for integration.

Not a fit

  • You need general ComfyUI installation or workflow development.
  • You require 24/7 support or a guaranteed SLA.
  • You plan to resell or redistribute the runtime image.

A maintained runtime, with a clear boundary.

Supported

  • Documented image starts on supported RunPod configurations.
  • Health checks and API Contract v1 work.
  • Released models, nodes and reference tests load.

Not included

  • Custom workflow development.
  • Frontend, database or billing integration.
  • RunPod outages, pricing or capacity.

Start with a technical fit review.

We review one existing workflow, your target GPU and the path to a first production job.

Book a technical fit review