ComfyUI beginner guide: your first image and common fixes

Start here

Intended reader: Beginners setting up ComfyUI who want to avoid red-node errors and dependency nightmares. Practical outcome: A working, crash-free image generation within 10 minutes and a repeatable mental model of data flow. An unfamiliar canvas, missing model files and a row of connected boxes can make the first image feel harder than it needs to be. Give yourself one small goal: run an official example, save the result and understand the path to the output.

Set a small, useful first goal

Your first goal is a saved image from a workflow you understand well enough to run again. It is not a giant graph that combines every popular extension. Choose a simple subject, such as a blue teapot on a kitchen shelf, and use it throughout your first session. Keeping the subject constant makes it easier to notice which setting changed the result. The checkpoints and folders below refer to the official basic example; other model families can require different loaders, encoders and files.

Load an official example and its required model

Comfy Org's first-generation guide explains loading example workflows from templates, JSON files or images containing workflow metadata. Follow the model link provided with the selected example. A workflow file describes the graph; it does not contain the model weights. For the basic checkpoint example, the guide places the model in the checkpoints folder under ComfyUI/models, then refreshes or restarts the application. Use the instructions for your actual installation because desktop and portable locations differ.

• Choose the example before downloading a collection of unrelated models.
• Wait for downloads to finish; confirm the expected filename.
• Select the matching file in the loader before running the graph.

Read the graph from inputs to output

In the documented basic text-to-image graph, the checkpoint supplies model components, text encoding produces conditioning, and a latent input gives the sampler an image space to work with. The sampled result is decoded before it becomes an image that can be saved. Following these connections is more useful than memorizing where boxes appear on the canvas: layouts can move while the data flow stays the same.

• Find the model input and identify the exact selected file.
• Locate the prompt input and the image-size settings.
• Follow the sampler output to decoding and saving.

Run once before you customize

Keep the example's remaining settings intact for the first attempt. Once an image appears, save the workflow and output together as a baseline. Write a tiny project note: what you expected, what appeared and whether anything failed. This is our recommended learning method, not a requirement imposed by ComfyUI. A plain but repeatable result is more useful at this stage than a polished image produced by a graph you cannot explain. Do not treat a successful basic example as proof that every larger model will fit your hardware.

Make a controlled comparison

Duplicate the working graph before editing. Keep the starting seed and other settings unchanged where the workflow allows it, then make one deliberate change to the prompt or another setting. Compare the two outputs at the same display size and write down the visible difference. This reduces ambiguity, but it does not promise identical results across different software versions or devices. Label the experiment with the change you made, not only a generic filename such as final-final.

• Baseline: the complete working example.
• Experiment 01: change only the subject or material description.
• Experiment 02: restore the baseline and change one other setting.
• Keep a short note explaining which version you prefer and why.

Missing model, incompatible file or memory error?

A missing entry in a model dropdown is a different problem from an out-of-memory failure after execution starts. Comfy Org's troubleshooting guide covers incorrect placement, incompatible models and insufficient resources. Read the first relevant error before installing more extensions or repeatedly downloading large files. Our triage approach is to identify the stage that failed, then verify the smallest set of assumptions at that stage.

• Missing file: compare the required filename and model folder with the example instructions.
• Incompatible workflow: check that the model family matches the loader and supporting components.
• Memory failure: return to the documented baseline, reduce unnecessary workload and check the model's requirements.
• Unclear failure: save the exact error and workflow version before changing the environment.

Add complexity only when it solves a problem

After the baseline works, define a reason for each new component. A LoRA might be part of a specific creative experiment; an upscaler might be needed for a final delivery size. Adding both while also changing the base model creates several possible explanations for any new failure. Keep a separate copy of the original graph and introduce one addition at a time. Before installing custom nodes, inspect their origin and maintenance instructions. They are software running in your environment, not passive artwork. Once your baseline image generation is reliable, advance to our ComfyUI FLUX and LoRA guide to customize model weights.

Save a project another person can understand

Create a project folder containing the workflow, a few representative outputs and a plain-language note. List the required models, relevant versions and the reason for unusual settings. Keep licensed assets and private reference images out of public workflow shares unless you have permission to distribute them. When asking for help, share the smallest reproducible graph and the actual error, rather than an entire private client folder. Good project notes also help you restart work after an update or a long break.

• Record model filenames and where their licenses are published.
• Keep the original baseline alongside later revisions.
• Describe the intended result and the current failure in separate sentences.
• Reopen the saved workflow to confirm your handoff is complete.

Try this next

Once the first workflow works, move on to a controlled FLUX and LoRA experiment. Read FLUX in ComfyUI: a practical workflow and LoRA checklist.