all systems operational
Get started
AI / media processing
apps › ai / media processing › comfyui

ComfyUI hosting on RTX PRO Blackwell GPUs.

Node-based Stable Diffusion and Flux on a dedicated GPU that's yours alone — up to 96 GB VRAM for the biggest models, NVMe for instant checkpoint and LoRA loads, and an API to wire ComfyUI into your pipeline. No per-image fees, no queue behind other tenants.

gpu
RTX PRO Blackwell
vram
up to 96 GB
tenancy
dedicated
deploy
2–6 h
/ why it needs real iron

Why a dedicated GPU beats a serverless endpoint.

96 GB VRAM, all yours

Flux, SDXL and large video models need real VRAM. An RTX PRO 6000 Blackwell gives you up to 96 GB with nobody else sharing it — load big checkpoints, high resolutions and long batches without out-of-memory errors.

No per-image bill

Serverless image APIs charge per generation and add up fast. A dedicated GPU is a flat monthly price — run it flat-out 24/7, render millions of images, the cost doesn't move.

NVMe for instant model loads

Checkpoints, LoRAs, VAEs and ControlNets are tens of gigabytes each. Enterprise NVMe loads and swaps them in seconds, so switching workflows doesn't stall your queue.

Your workflows, your API

ComfyUI's node graphs and REST/websocket API run on a box you fully control — wire it into your app, batch jobs from a queue, or expose a private UI. Unmetered bandwidth pulls models and pushes outputs for free.

/ recommended configurations

Sized for ComfyUI.

Starter / Production / High-load — with capacity in the metrics that matter for this app.

KyivAmsterdamWarsawMiami
Creator
SDXL / Flux · 32 GB VRAM · single artist
  • CPUEPYC + RTX PRO 4500
  • RAM128 GB
  • Storage4 TB NVMe
  • Port10 Gbit/s
  • Trafficunmetered
$550/mo
Deploy Creator →
Recommended
Studio
big models · 96 GB VRAM · batch pipelines
  • CPUEPYC + RTX PRO 6000
  • RAM256 GB
  • Storage8 TB NVMe
  • Port10 Gbit/s
  • Trafficunmetered
$990/mo
Deploy Studio →
Farm
multi-GPU · high-throughput generation API
  • CPU2× RTX PRO 6000
  • RAM512 GB
  • Storage16 TB NVMe
  • Port25 Gbit/s
  • Trafficunmetered
$1890/mo
Deploy Farm →
/ what’s included

On by default.

ComfyUI pre-installed

ComfyUI with CUDA, PyTorch and the common custom nodes set up, behind an HTTPS reverse proxy — plus the API enabled for programmatic jobs.

Dedicated RTX PRO GPU

A whole RTX PRO 4500 or 6000 Blackwell — not a shared slice — with up to 96 GB VRAM for the largest image and video models.

NVMe model storage

Multi-terabyte NVMe for checkpoints, LoRAs, VAEs and ControlNets, so model loads and workflow swaps happen in seconds.

Unmetered port

10 / 25 Gbit/s at flat pricing — pull huge models and push rendered outputs with no per-GB egress.

API-ready

ComfyUI's REST/websocket API exposed for batch jobs and app integration; queue work from your own backend.

Root access

Full SSH/root — install any custom node, model or extension, pin CUDA and driver versions, script whatever you need.

ComfyUI hosting — dedicated RTX PRO GPUs for Stable Diffusion & Flux

ComfyUI is the node-based front end power users reach for to run Stable Diffusion, SDXL, Flux and video models — visual workflows, precise control, and a REST/websocket API to automate everything. What it demands is a real GPU with a lot of VRAM and fast local storage for the pile of multi-gigabyte checkpoints and LoRAs a serious workflow touches.

Hostfory runs ComfyUI on dedicated RTX PRO Blackwell GPUs — not a shared serverless slice — with up to 96 GB of VRAM, so the biggest models and high resolutions fit without out-of-memory errors, and nobody else is queued behind you. Because it's a flat monthly server rather than a per-image endpoint, you can run it around the clock and render at volume without a metered bill, and the unmetered port pulls models and pushes outputs for free. Enterprise NVMe keeps checkpoint and LoRA loads instant so switching workflows never stalls the queue.

You get full root access to install any custom node, model or extension and pin your CUDA and driver versions, plus the API wired up for batch jobs and app integration. Servers provision in a few hours with ComfyUI, CUDA and PyTorch ready — pair it with the Whisper / ffmpeg GPU box for a full AI-media pipeline.

GPURTX PRO 4500 / 6000 Blackwell
VRAMup to 96 GB
StackComfyUI · CUDA · PyTorch
Modelsmulti-TB NVMe · checkpoints / LoRAs
Accessroot · API enabled
LocationsAmsterdam · Warsaw · Kyiv · Miami
/ faq

ComfyUI questions, answered.

Is the GPU dedicated or shared?
Dedicated — you get a whole RTX PRO 4500 or 6000 Blackwell, not a time-sliced share. All the VRAM (up to 96 GB) and all the compute are yours, with no queue behind other tenants.
How is this cheaper than a serverless image API?
Serverless endpoints bill per generation. A dedicated GPU is a flat monthly price, so if you generate at any real volume — batch jobs, a product feature, constant experimentation — running your own box is dramatically cheaper and the cost is predictable.
Which models can it run?
SD 1.5, SDXL, Flux, and large video and upscaling models. The RTX PRO 6000 with 96 GB VRAM handles the biggest checkpoints, high resolutions and long batches; the 4500 is sized for SDXL/Flux single-artist work.
Can I use ComfyUI's API to automate jobs?
Yes — the REST/websocket API is enabled so you can queue workflows from your own backend, build a product feature on top, or run scheduled batch renders. Full root access lets you script the rest.
Can I install custom nodes and models?
Yes — you have root. Install any custom node, checkpoint, LoRA or extension, and pin CUDA and driver versions to match your workflow.
Is the bandwidth metered?
No — the port is unmetered, so pulling tens of gigabytes of models and pushing rendered outputs costs nothing extra on top of the flat monthly price.
How long does setup take?
A few hours — the server provisions with ComfyUI, CUDA and PyTorch installed and the API ready behind HTTPS.

Deploy ComfyUI today.

In-stock configs provision in 1–3 hours. Migration from another host is on us.

Chat with us@hostfory