ComfyUI
Understand one image-generation workflow by tracing its input, model, and output nodes.
- Suitable for
- Visual creators who want to understand and modify a generation pipeline rather than use a one-box prompt screen.
- Do not start here if
- Not ideal for weak computers, low storage, or anyone who wants a one-click first image with no setup.
- Visible result
- One official or built-in workflow opens, every required node is present, and one test image is saved with its workflow provenance.
- Time
- 90 min
- Device
- A supported computer; a capable GPU is strongly preferred
- Account
- Base local use may not need an account; model downloads can have separate terms
- Cost
- Software is open source; hardware, storage, models, or cloud GPU can cost money
- Data
- Substantial free storage; exact model sizes must be checked before download
- HAACC check
- Editorially reviewed
Not run by HAACC
Before you start
A supported computer; a capable GPU is strongly preferred
Base local use may not need an account; model downloads can have separate terms
Software is open source; hardware, storage, models, or cloud GPU can cost money
Substantial free storage; exact model sizes must be checked before download
Patience to trace nodes and avoid unreviewed custom extensions
The path
- 01
Check GPU and storage
Record GPU memory, system memory, and free disk space. Pick the smallest current official workflow and model that fits those limits.
Continue only whenRequired model files and their sizes are known before download. - 02
Start without custom nodes
Use only the current official installation path and built-in nodes. Do not install extension managers or workflow packs for the first result.
Continue only whenThe interface opens with no missing custom-node warning. - 03
Trace before generating
Identify the prompt input, model loader, sampler, decoder, and save node. Write one sentence describing what each does before pressing run.
Continue only whenYou can follow one connection from prompt to saved image. - 04
Render one small test
Use a harmless prompt and conservative image size. Save the image together with the workflow metadata or exported workflow file.
Continue only whenOne image is saved and the workflow that produced it can be reopened.
One image is produced by a traceable built-in workflow with no unreviewed custom nodes.
Evidence boundary
HAACC reviewed the public source and shaped this path, but did not run the project on a clean machine. Treat every install step as unverified.
What HAACC checked
- Repository identity, snapshot metrics, declared GPL-3.0 license, and core node-graph concept
- A minimal first path that excludes unreviewed custom nodes
What remains unchecked
- Installation, model compatibility, performance, and removal on any hardware
- Licenses and permitted uses of downloaded models and generated assets
Stop and read
Custom nodes execute third-party code
Do not install unknown custom nodes for this Guide. Popularity is not a security review.
Model and image rights vary
Check the exact model license, training-data policy, subject rights, and intended commercial use before publishing.
When something fails
Out-of-memory or system freeze
Stop the queue, reduce image size or choose a smaller official model. Do not repeatedly force the same oversized workflow.
The workflow reports missing custom nodes
Do not install unknown nodes. Return to a built-in official workflow for the first result.
Stop the application, remove downloaded models and generated trial files you do not need, uninstall with the current official method, and confirm storage is reclaimed.
Turn it into your project
One visual system, three prompts
Keep the workflow fixed and document how three prompt changes affect the result.
Provenance-first gallery
Publish each image with model, workflow, settings, rights notes, and known limits.
Source snapshot
- Repository
- Comfy-Org/ComfyUI
- Captured
- 2026-07-20
- Stars at capture
- 121,410
- Forks at capture
- 14,299
- GitHub license field
- GPL-3.0
- Pinned revision
- Not pinned