Is this difference caused by:
a ComfyUI version difference, or
the use of custom (external) nodes?
maybe both or similar…?
Why you don’t see Apply ControlNet (Advanced) in ComfyUI Cloud
The node was renamed
In current ComfyUI, “Apply ControlNet (Advanced)” was renamed to “Apply ControlNet.” (comfyui-wiki.com)
At the same time, the very early node that used to be called “Apply ControlNet” was renamed to Apply ControlNet (Old) and is typically hidden from search (kept only for backward compatibility with older workflows). (comfyui-wiki.com)
So the simplest explanation for your cloud environment is:
- Your Cloud image is using a newer ComfyUI where the tutorial’s node name no longer exists.
- The tutorial is using an older naming era (or a workflow recorded before the rename).
Why the tutorial’s node has no VAE port, but yours does
What changed
ComfyUI introduced a “new Apply ControlNet” that can handle both:
- SD1.5 / SDXL-style ControlNets (often fine without VAE),
- and newer model families / ControlNet pipelines (which may require VAE).
A ComfyUI GitHub support issue about the “new apply controlnet node” explicitly summarizes the intent: the VAE input is optional so it can work for SD1.5/SDXL without VAE and for SD3/Flux with VAE. (GitHub)
Separately, there are real reports that ComfyUI updates changed an input that used to be a FLOAT into a VAE-typed input, which broke older graphs—this matches the kind of “same node, different port types” mismatch you’re noticing. (GitHub)
What the VAE port is for (beginner-friendly background)
- Stable Diffusion pipelines frequently operate in latent space.
- The VAE is the component that converts image → latent (encode) and latent → image (decode).
- Some ControlNet implementations need access to that encoding/latent representation, so ComfyUI exposes a VAE input to support those cases (or make them work reliably across model families). (GitHub)
Practical guidance
- If your workflow runs fine: you can usually leave VAE unconnected (because it’s optional in many setups). (GitHub)
- If you see an error that a ControlNet needs a VAE: connect the VAE from your checkpoint loader / VAE loader.
Is this caused by a ComfyUI version difference, or custom nodes?
For your specific difference (Advanced missing + VAE port present)
This is overwhelmingly explained by ComfyUI core changes (rename + updated node definition), not by a missing custom node pack. (comfyui-wiki.com)
Why tutorials often diverge
YouTube workflows frequently mix:
- older ComfyUI builds,
- older workflow JSONs,
- and custom node packs.
So two people can both say “Apply ControlNet (Advanced)” but be referring to different eras (or even different node implementations).
Where “custom nodes” fit in, and whether they’re “better”
What custom nodes are
Custom nodes are third-party extensions that add new nodes / features beyond ComfyUI core. (ComfyUI)
Are they “better”?
Not inherently. Think of it like this:
Core nodes
- Usually more stable across updates (maintained together with ComfyUI).
- Best for learning fundamentals.
Custom nodes
- Can unlock capabilities ComfyUI core doesn’t include (e.g., many preprocessors, specialized workflows).
- But they can break when ComfyUI updates (version mismatch is common), and they add dependency/security risk. (ComfyUI)
ComfyUI’s own install docs explicitly warn that custom nodes listed in managers/registries aren’t automatically “safe,” and you should install only trusted/popular ones. (ComfyUI)
Can you install custom nodes in ComfyUI Cloud?
This depends on what you mean by “Cloud.”
If you mean official Comfy Cloud (comfy.org)
Official docs describe Cloud as: “Use pre-installed custom nodes and never worry about compatibility issues,” contrasted with self-hosted/local where you can install any custom node but must manage it yourself. (ComfyUI)
Implication: official Comfy Cloud is typically curated/limited (you don’t necessarily get arbitrary GitHub installs like local).
If you mean a third-party hosted ComfyUI (RunComfy / ThinkDiffusion / your own VM, etc.)
Capabilities vary by provider:
- Some hosts support installing nodes via ComfyUI-Manager and persist them.
- Others lock the environment and only allow preinstalled sets.
(For local/self-hosted, ComfyUI documents standard installation methods: clone into custom_nodes/ + install dependencies.) (ComfyUI)
What you should do to follow the tutorial successfully
-
Use “Apply ControlNet” as the modern replacement for the tutorial’s “Apply ControlNet (Advanced).” (comfyui-wiki.com)
-
Ignore the VAE port unless you need it (it’s meant to be optional in many cases). (GitHub)
-
If the tutorial uses a deprecated node, you can enable deprecated nodes:
- Settings → Comfy → Node → “Show deprecated nodes in search” (ComfyUI)
But the official tutorial recommends using the newer node. (ComfyUI)
Quick diagnostic for your exact cloud environment
- If your UI is Comfy Cloud (comfy.org), assume a curated node set (preinstalled custom nodes). (ComfyUI)
- If your host provides ComfyUI-Manager, you may be able to install custom nodes; if not, it’s likely restricted.