Reading someone else's workflow

A method for understanding a workflow you downloaded, before you run it.

The method

Read the notes. Template and workflow authors put the list of model files, folders and settings in Note or Markdown Note nodes.
Find the outputs. Look for Save Image, Save Video, Preview Image. Everything that matters leads to them; branches that lead nowhere do not run.
Walk back from each output to the sampler that feeds it. The sampler is the heart: look at its model, positive, negative and latent_image inputs.
Find the loaders. Which checkpoint or diffusion model, which text encoders, which VAE, which LoRAs. That tells you the family, and the family tells you what files you need. See Model families.
Find the start. An empty latent (text to image), a Load Image and VAE Encode (image to image), a mask (inpainting), a video start node.
Read the prompts and the sampler settings: steps, cfg, sampler, scheduler, denoise, seed.
Look for the unusual: several samplers (two passes, refining, upscaling), ControlNet, reference images, subgraphs, Set/Get nodes that hide links, bypassed or muted parts.
Check what is missing: nodes from packs you do not have, model files you do not have.
Check for paid or risky parts: Partner (API) nodes that cost credits and send data out; packs you do not know.

Running a big workflow before checking its memory needs: start with a smaller size or length.
Installing every missing pack without looking at what it is: see Node packs.
Replacing a missing model with one of another family: the graph runs but the result is noise.

File ▸ Open workflow… shows after import what was found and what your ComfyUI lacks.
Guided tour (Graph ▸ Guided tour) walks through the graph node by node with explanations.
The Explain tab describes whatever you select; Check lists missing nodes, missing models and families that do not match.
Ask mode lets the assistant answer questions about the graph on the canvas.
The lesson Reading someone else's workflow practises it.