What Is ComfyUI? Is It Free, How to Install and Use It, How Much VRAM It Needs and What Comfy Cloud Costs

ComfyUI is an open-source program that generates images, video, audio and 3D content with AI. Each step is placed on screen as a node (a box): load a model, write text, generate an image, save. You connect the nodes with lines; this connected structure is called a workflow. That way you can reuse the same settings again and again, or open and run someone else's workflow.

What happens locally: The program and model files stay on your computer, and generation runs on your graphics card. In the cloud, the same interface runs on the company's servers.

Who it's for: teams and content creators who want to make product images, design experiments and short videos, and who want full control over models and settings.

ComfyUI's advanced features: App Mode, which turns a saved workflow into a simple interface, and a local API that connects workflows to other apps. Both are listed as features in the README; setup details aren't covered in this guide.

The official README recommends the desktop app (Comfy Desktop) for beginners. The portable version and the manual install are more technical; the desktop app follows the stable release, while the portable version and manual install give you the latest code.

Source: ComfyUI README, docs.comfy.org installation pages, Comfy Cloud documentation. According to the docs, the desktop app is also available on Linux; the README lists only Windows and macOS.

For the question of which graphics card to buy, the README points to the “Which GPU should I buy for ComfyUI” page on the GitHub wiki.

!“The software is free” doesn't mean “no computer costs”: the graphics card, electricity and disk space (model files are large) are on you. This guide doesn't give a server or graphics card cost, and neither do the official sources.

1. 1Open a template. Pick a ready-made workflow from the template icon in the left sidebar. Starting from a template is easiest, rather than building nodes from scratch.
2. 2Put the model file in place. Locally, you need the model files the template requires. According to the README, main model (checkpoint) files go in the models/checkpoints folder and VAE files in models/vae. This step doesn't exist in the cloud.
3. 3Change your text and image. Describe the image you want in the CLIP Text Encode node; in templates that take an image, add yours to the Load Image node.
4. 4Run it. The run icon or Ctrl + Enter. If there's an error, read the details with Show report.
5. 5Save. Right-click the generated image and choose Save image; for video, use Download in the player's three-dot menu.

If you drag a generated PNG file onto the ComfyUI page, the workflow that produced that image comes back, along with its seed values (README).

ComfyUI isn't a model; it's an engine that runs models. The README lists supported models by category and says this is a representative list. For the current, complete list, see the tagged templates on comfy.org/workflows. Checked on October 14, 2026.

A custom node is a community-written extension that adds features to ComfyUI. ComfyUI-Manager is the extension that finds, installs, updates and manages them. According to the README, for a manual install you first run pip install -r manager_requirements.txt, then start ComfyUI with python main.py --enable-manager.

!Custom nodes are third-party code and run on your computer. The official Registry standards ban the use of eval and exec, installing packages at runtime, and obfuscated code; the stated reason is that a node passing user input to eval/exec could open the door to attacks such as keyloggers or ransomware. These rules are a requirement for publishing to the Registry: there's no guarantee that a node you copied by hand from GitHub meets them.

• Prefer actively maintained nodes published on the Comfy Registry; don't install every missing node a workflow from an unknown source asks for.
• If a workflow comes from someone else, look at its list of nodes and check where each one comes from.
• Commits other than the stable release tag may be unstable and can break many custom nodes (README).
• There's no official Docker image; community images aren't supported by Comfy Org (docs).

Source: ComfyUI README, Comfy Registry Node Quality and Security Standards, system requirements FAQ. This isn't a “safe” stamp; it only summarizes what the official documentation says.

Locally, the software is free. There's no subscription or quota for ComfyUI installed on your own computer. Costs come up in two places: Comfy Cloud (cloud GPUs) and Partner Nodes (closed-source models). Both are paid with the same Comfy Credits.

• What credits are spent on: active GPU time used while a workflow is running, and Partner Nodes. Sitting idle or building a workflow doesn't use credits.
• Example: according to the pricing page, a 5-second video with default settings costs about 11 credits. Heavy models, high resolution and Partner Nodes use more.
• Rollover: plan credits reset at the end of the billing period and don't roll over. Credits you buy separately (top-up) are valid for 1 year.
• Free trial: the pricing page lists 5 free runs for new users; the subscription documentation says a paid plan is required to run workflows. The two conflict: check what you see on the sign-up screen.

The official pages make no statement specific to KVKK (Turkey's Personal Data Protection Law). Read the privacy policy before processing images that contain personal data in the cloud.

• For every model you'll use in client work, read that model's license (does it allow commercial use?) on its own page.
• Locally, commercial licenses for some models are sold separately: the pricing page says Comfy is the official reseller for MiniMax models and that the Professional license starts at $5,000 a month; Cloud is said to include commercial use.
• For models in Partner Nodes, the providers' commercial license and data information is on a separate verification page.

Source: GitHub license information, Comfy Desktop documentation, Comfy Cloud pricing page. This isn't legal advice.

Source: docs.comfy.org troubleshooting, custom node troubleshooting, system requirements and desktop installation pages.

ComfyUIAn open-source engine where you connect and control models with nodes; free locally, but installation and learning are on you.MidjourneyA ready-made image generation app you use without installing or managing models (details in its own guide).CanvaOffers image generation inside a design tool; no model choice and no workflow building.

The official sources don't compare the tools. Try the same job in both.