ComfyUI won a particular argument. If you want control over image and video generation rather than a text box and a hopeful click, a node graph is the honest interface, because the pipeline really does have stages and they really do need wiring. The cost of that honesty is that a beginner opens ComfyUI and sees a circuit diagram. The official documentation is the main thing standing between that beginner and giving up, and it has come a long way.
Installation is covered properly across platforms, the interface and node system are explained, and the tutorials are graded by difficulty with a basic text to image walkthrough as the entry point, then further material on video, audio and 3D generation. The built in nodes reference is the part I use most, because once you are actually building something the question is never how do I start, it is what does this sampler parameter do and why did changing it destroy my image. Where it falls short is the difference between documentation and teaching. Documentation tells you what a node does.
Teaching tells you why the graph is arranged the way it is, what each stage of a diffusion pipeline is for, and how to reason about a result that came out wrong. The tutorials give you working workflows, and following one successfully is not the same as understanding it, which is how you end up with a large collection of downloaded workflows you cannot debug. Anyone serious should spend a few hours learning what diffusion actually does before or alongside this, because the graph becomes almost self explanatory once you know that a latent gets noised, denoised under text conditioning, and decoded. Churn is the second problem and it is structural.
ComfyUI develops quickly, the custom node ecosystem develops even faster, and much of what people actually use in practice lives in community nodes that the official documentation cannot cover. That means screenshots drift, menu paths change and the gap between the documented tool and the tool people are using is permanently open. Troubleshooting reflects this, since a real problem often ends with you reading a GitHub issue thread or asking on Discord, which works and is not documentation. Hardware is worth mentioning too.
Running locally is a genuine advantage, because there is no subscription and no per image cost and no content filter deciding what you may generate, and it also means the requirements are yours to meet. On a modest GPU you will be waiting, and some of the video and larger model tutorials will be out of reach entirely. My 3.6 is a rating of solid, free, much improved product documentation for a tool with a hard learning curve. It does the documentation job competently.
It does not do the teaching job, and the reason so many people find ComfyUI impenetrable is that they came here first, when the concepts they were missing were never this site's responsibility to supply.