ComfyUI Templates in 2026: 660 Built In, and Half Need the Cloud

Quick answer

• ComfyUI ships 660 workflow templates in the version current at the time of writing — far more than the handful most guides show you.
• 🔴 368 of them (56%) are cloud API templates. They send your prompt, and often your image, to Google, Grok, MiniMax, Recraft or similar. On a tool people install specifically to run things on their own machine, the majority of built-in templates are not local.
• 292 are local. Of the 272 that declare a size, the median model download is 23.4 GB and the largest is 103.8 GB. Pulling every one would cost about 7.2 TB.
• Only 13 local templates download under 2 GB — and those are the sensible place to start.
• Node Basics is the only group that is 100% local. Vector is 100% cloud.

The figures below were computed from the template catalogue itself (`templates/index.json` in the
Comfy-Org/workflow_templates repository, MIT-licensed), read on 28 September 2026. They are counts, not estimates, and you can reproduce them.

What a template actually is

A template is a pre-built node graph you load instead of wiring one yourself. It ships as a package — comfyui-workflow-templates on PyPI, at v0.11.70 (25 September 2026), with the repository one version ahead at v0.11.71 — so it updates independently of ComfyUI itself.

The important mechanic is what happens on load. Per ComfyUI’s own documentation, “Each workflow template embeds links to the required models”, and “When loading a template, ComfyUI automatically checks whether all required model files exist. If anything is missing, it will prompt you to download the models.”

That download prompt is where the disk cost lands, and it is why the size figures below matter more than the template count.

More than half the templates are not local

This is the finding that changes how you use the browser, and it is not mentioned in the template picker itself.

The pattern is worth reading carefully. The Applied groups — Product & Ads, Brand & Design, Vector — are almost entirely cloud, between them offering 2 local templates out of 38. The Node Basics group is the only one with no API templates at all, which makes it the honest starting point if running locally is the reason you installed ComfyUI. If what you actually want is a canvas rather than a node graph, ComfyUI may be the wrong tool entirely — see Stable Diffusion image editors in 2026.

The catalogue marks these itself: 367 templates carry openSource: false, and the same set (bar one) is tagged API or named with an api_ prefix. So the information is there — it is just not surfaced as a filter when you are choosing.

How to tell before you click

API templates are named with an api_ prefix in the catalogue and tagged API in the template browser. They also require an API key and will bill you per generation. If a template names Nano Banana Pro, Gemini 3 Pro Image, Grok, MiniMax or Recraft, it is a hosted model, not a local one.

What the local templates cost in disk

Of the 292 local templates, 272 declare a model download size. Twenty do not, and are excluded rather than guessed at.

Median 23.4 GB, mean 26.6 GB, smallest 0.11 GB, largest 103.8 GB. The whole local set is roughly 7.2 TB. Half of all local templates want more than 23 GB on disk before they will run once, and only 63 of 272 come in under 12 GB.

These are download figures, not VRAM. A 24 GB download does not mean you need a 24 GB card — quantization and offloading change what actually has to be resident. For the memory side, our image generation VRAM calculator computes peak usage per model and precision, and the guide to running Stable Diffusion locally works through the tiers.

The 13 templates that download under 2 GB

If you want something that works today on an ordinary machine, start here. Note that most are utilities rather than generators — which is the honest shape of the accessible end of this catalogue.

The gotcha in missing-model detection

Worth knowing before you spend an evening on a failed download. ComfyUI’s documentation states that “In the current version, missing-file detection only checks whether there is a file with the same name in the corresponding top-level directory.”

So a file you already have, filed in a subfolder, will not be detected — and a file with the right name but the wrong contents will be. If a template keeps prompting for a model you are sure you downloaded, check it sits directly in the expected folder (for example ComfyUI/models/diffusion_models/) rather than nested inside it. If you keep models in a shared location, extra_model_paths.yaml is the mechanism, which we cover in checkpoints, LoRAs and VAEs explained.

How to pick one

• New to ComfyUI and running locally: Node Basics — 7 templates, all local, no API keys.
• Want a result today on modest hardware: one of the 13 sub-2 GB utilities above. Background removal and depth estimation both work well and download in seconds.
• Generating images locally: the Image group has the most local options of any group (107), so it is the best-stocked place to look.
• Video: expect the largest downloads. Video is 58% cloud, and the local remainder is where most of the 50 GB-plus templates live. Our local AI video generation guide covers what those models need.
• Anything commercial: check the model licence separately from the template. The template being MIT says nothing about the weights it pulls — see our Flux licence guide for how sharply that can differ.

FAQ

How many templates does ComfyUI have?

660 in the catalogue current at the end of September 2026, across 12 groups. The count changes often — the templates package shipped two releases in the three days before this was written.

Are ComfyUI templates free?

The templates are, under the MIT licence. But 368 of the 660 call a paid cloud API and need a key, and the 292 local ones require model downloads averaging 26.6 GB. Free to load is not the same as free to run.

Which ComfyUI template should a beginner use?

One from Node Basics, which is the only group with no API templates. If you want a finished result rather than a lesson, Remove Background: BiRefNet downloads 0.43 GB and does something useful immediately.

Why does ComfyUI keep asking me to download a model I already have?

Detection only looks for a matching filename in the top-level model directory, not in subfolders. Move the file directly into the expected folder, or point ComfyUI at your existing library with extra_model_paths.yaml.

Sources

• Template catalogue and counts computed from templates/index.json (github.com/Comfy-Org/workflow_templates, MIT), read 28 September 2026
• Template loading, model storage and missing-file detection (docs.comfy.org)
• Package version and release date (pypi.org)

Updated 28 September 2026: all counts and download sizes computed from the template catalogue on this date. ComfyUI ships template updates frequently, so treat the totals as a snapshot and re-run the numbers if precision matters.