Troubleshooting Common ComfyUI Errors

Errors Are Part of the Workflow

ComfyUI is powerful precisely because it exposes the generation process directly — and that means when something goes wrong, you see the failure, not a vague "generation failed" message. The error system is actually more informative than most AI tools. Once you know where to look and what the common errors mean, diagnosing problems takes under a minute for most situations.

This lesson covers the errors you will encounter most often and how to resolve each one. The pattern is always the same: read the error, match it to a cause, apply the fix.

Reading the Console Panel

When a generation fails, the Console panel in the bottom toolbar is where the error information appears. Click Console to open it. You will see a log of ComfyUI's activity, with errors highlighted in red.

The most important thing to understand about ComfyUI error logs: there is usually a cascade — several error messages appear, one after another, because each step in the pipeline fails after the previous one does. Focus on the first red error in the log, not the last. Subsequent errors are usually symptoms of the first failure, not independent problems.

ComfyUI also shows error information directly in the interface: a red border or overlay on the failed node, and a pop-up message at the top of the canvas. The pop-up is usually a shortened version of the error. The Console has the full message.

CUDA Out of Memory (OOM)

Error message: torch.cuda.OutOfMemoryError: CUDA out of memory or RuntimeError: CUDA out of memory.

Cause: The GPU does not have enough VRAM to hold the model and the current generation parameters simultaneously.

Fixes, in order of how much they help:

• Reduce resolution: In the Empty Latent Image node, lower the width and height. Halving both dimensions reduces VRAM usage by approximately 4x. For SD 1.5, try 512x512. For SDXL, try 768x768. For FLUX.1, try 768x1024.
• Reduce batch size: In the Empty Latent Image node, set batch_size to 1 if it is higher.
• Enable CPU offloading: Restart ComfyUI with the flag --lowvram (for 4–6 GB VRAM) or --novram (for less than 4 GB). Add these flags to your startup command: python main.py --lowvram. On Desktop, add them in the launcher settings.
• Free VRAM from other applications: Close other GPU-intensive programs (games, video editors, other AI tools) and restart ComfyUI.
• Use a smaller model: Some models have quantized variants (.gguf or fp8 precision) that use significantly less VRAM.

Missing Node Errors (Red Boxes)

Error message: A node on the canvas turns red/pink with the message "Custom nodes not installed" or a similar warning.

Cause: The workflow requires a custom node package that is not installed in your ComfyUI setup.

Fix:

• Open ComfyUI-Manager (Manager button in the top bar).
• Click Install Missing Custom Nodes.
• Review the list of missing packages. Click Install next to each one (or use Select All).
• After installation completes, restart ComfyUI.
• If the red box persists after restart, check whether the package name in Manager matches what the workflow expects — some packages have been renamed or superseded by newer versions.

Model Not Found

Error message: ValueError: model not found or the node shows a red warning and the model dropdown displays a name that is not in the list.

Cause: The workflow references a model file that is not in your models folder.

Fix:

• In the affected node (Load Checkpoint, Load LoRA, etc.), click the model dropdown and select any model you have installed. If the dropdown is empty, you do not have any models of that type installed yet.
• Download the specified model (or an equivalent one) and place it in the correct subfolder as described in Lesson 6 and 11.
• Press R on the canvas to refresh the model list, then re-select from the dropdown.
• Check for path and case issues: ComfyUI on Linux is case-sensitive. A file named DreamShaper_8.safetensors will not be found if the workflow specifies dreamshaper_8.safetensors.

Broken Connections and Type Errors

Error message: A connection refuses to form (wire snaps back), or the generation fails with a type mismatch message like expected type IMAGE, got LATENT.

Cause: An output of one type is connected to (or attempted to be connected to) an input of a different type.

Fix: Review the node connection chain. The data types must match at every connection point. If you need to convert between types (e.g., IMAGE to LATENT), add the appropriate converter node (VAE Encode converts IMAGE to LATENT; VAE Decode converts LATENT to IMAGE). Check the colors of the output and input slots — if they do not match, the connection is not valid.

Queue Stuck or Not Processing

Symptom: You pressed Ctrl+Enter but nothing happens — the queue counter shows a pending job but generation never starts.

Common causes and fixes:

• Workflow has a validation error: ComfyUI will not start generation if it detects a configuration problem. Check the Console for a validation message, fix the highlighted node.
• Previous generation crashed and left the system in a bad state: Click Clear Queue in the top bar, then try running again.
• ComfyUI process is frozen: Stop ComfyUI (Ctrl+C in the terminal), restart it, and reload the workflow.

Updating ComfyUI

Keeping ComfyUI updated ensures you have the latest model support, bug fixes, and performance improvements. For each installation method:

• Desktop app: The app checks for updates automatically. A notification appears when an update is available. Click Update.
• Portable: Download the new archive from the GitHub releases page and copy your models/ and custom_nodes/ folders into the new installation.
• Manual install: Activate your conda environment, navigate to the ComfyUI folder, run git pull, then run pip install -r requirements.txt to update dependencies. Restart ComfyUI.

After updating ComfyUI, run Update All in ComfyUI-Manager to update your custom nodes as well — ComfyUI API changes can break older custom nodes that have not been updated to match.

With this lesson complete, you have everything you need to install, navigate, run, organize, and troubleshoot ComfyUI. The next track in this series — ComfyUI Image Generation — picks up from here and covers the powerful generation techniques: FLUX.1 workflows, ControlNet, LoRA, IPAdapter, and building production image pipelines.