Comfyui Keeps Reconnecting: Easy Fixes That Actually Work

Comfyui Keeps Reconnecting is one of those errors that can turn a simple image-generation session into a confusing mess. You open your workflow, press queue, and suddenly the browser shows “reconnecting,” “connection lost,” or “failed to fetch.” In many cases, this does not mean your whole ComfyUI installation is broken. It usually means the browser has lost contact with the local ComfyUI server, the backend has restarted, a custom node has crashed, or the workflow is demanding more memory than your system can handle.

ComfyUI works through a local server and a browser-based interface. When the backend stops, restarts, freezes, or cannot respond properly, the frontend tries to reconnect. That is why the error often appears during heavy model loading, after updating extensions, when using custom nodes, or when running large workflows such as FLUX, SDXL, video, or GGUF-based models.

Why Comfyui Keeps Reconnecting

When Comfyui Keeps Reconnecting, the first thing to understand is that the browser is usually not the real problem. The “reconnecting” message is often a symptom. The real cause is usually happening in the terminal, command prompt, Python environment, custom nodes, GPU memory, or network connection between the browser and the ComfyUI server.

ComfyUI is a powerful modular node-based AI creation tool, and its GitHub project describes it as an engine for visual AI workflows across images, video, 3D, audio, and more. That flexibility also means many things can affect stability: models, custom nodes, extensions, Python packages, GPU drivers, browser cache, and local server settings.

A reconnect loop usually happens for one of these reasons:

The ComfyUI backend has crashed.

The browser cannot reach 127.0.0.1:8188.

A custom node is causing an error.

A workflow is using too much VRAM or RAM.

ComfyUI Manager applied changes and restarted the server.

A proxy, cloud setup, or remote server is interrupting WebSocket communication.

A browser extension, VPN, firewall, or antivirus is blocking the connection.

The good news is that most of these issues can be fixed without reinstalling everything.

Check the Terminal Before Changing Anything

The fastest way to diagnose the problem is to look at the terminal or command prompt where ComfyUI is running. If the browser says “reconnecting,” but the terminal has closed, frozen, or shows a Python traceback, the backend has failed.

Do not start by changing random settings. First, scroll through the terminal and look for clear messages like “ModuleNotFoundError,” “CUDA out of memory,” “ImportError,” “Failed to load,” or “custom node error.”

For example, some user-reported ComfyUI failures show errors connected to outdated custom node APIs or node-loading failures. In one GitHub issue, the logs mention deprecated legacy API imports and errors while retrieving node information, which points toward extension compatibility rather than a basic browser issue.

If the terminal gives you an error, copy the last 20–40 lines. That section usually tells you which model, extension, or Python package caused the reconnect.

Restart ComfyUI the Right Way

A normal page refresh may not fix this error because the browser is only the frontend. You need to restart the backend too.

Close the ComfyUI browser tab. Then stop the ComfyUI terminal window completely. On Windows, close the command prompt or press Ctrl + C if it is still active. After that, start ComfyUI again using your normal launch file, such as run_nvidia_gpu.bat, run_cpu.bat, or your Python command.

Once the server is fully loaded, reopen:

http://127.0.0.1:8188

If the page loads normally, queue a very simple workflow first. Do not immediately run the same heavy workflow that triggered the reconnect. A small test helps you confirm whether ComfyUI itself works.

Update ComfyUI and the Frontend

If Comfyui Keeps Reconnecting after a recent model or workflow change, your installation may be outdated. ComfyUI changes quickly, and custom nodes often depend on newer frontend or backend behavior.

For a Git-based installation, you can usually update from the ComfyUI folder with:

git pull

Then update Python dependencies if needed:

pip install -r requirements.txt

If you use ComfyUI Manager, open Manager and update ComfyUI, custom nodes, and dependencies from there. ComfyUI Manager is designed to help install and manage custom nodes, and its official repository includes installation steps for adding it into the custom_nodes folder.

After updating, restart ComfyUI completely. Do not rely only on a browser refresh.

Be Careful After Using “Apply Changes” in ComfyUI Manager

Sometimes the reconnect message appears right after updating nodes or applying changes. That can be normal because ComfyUI may restart after changes.

However, a recent ComfyUI frontend issue described a situation where “Apply Changes” could trigger a restart, the WebSocket reconnects, but the HTTP runtime and node definitions may not be ready yet on slower or proxied setups. In simple terms, the interface may reconnect before the backend is fully ready.

If this happens, wait until the terminal clearly shows that the server is ready. Then refresh the browser manually. If the workflow still fails, restart ComfyUI again and test with a blank or simple workflow.

Disable Custom Nodes to Find the Problem

Custom nodes are one of the most common reasons ComfyUI starts reconnecting. A node may be outdated, missing a dependency, using an old API, or conflicting with another extension.

The official ComfyUI troubleshooting documentation recommends a binary-search style method for custom node problems. That means disabling half of the custom nodes, testing, and then narrowing the issue down step by step until you find the broken extension.

Here is the practical version.

Go to your ComfyUI folder.

Open:

ComfyUI/custom_nodes

Move half of the custom node folders into a temporary backup folder outside ComfyUI.

Restart ComfyUI.

Test the same workflow.

If the reconnecting stops, one of the removed nodes is likely responsible. If it continues, the issue may be in the remaining nodes or somewhere else.

Keep narrowing the folders until you find the exact custom node. Once found, update it, reinstall its requirements, or remove it if it is no longer maintained.

Fix Missing Python Dependencies

Sometimes ComfyUI reconnects because a node needs a package that is not installed. This often happens after installing a new custom node manually.

Open the custom node folder and look for a file named:

requirements.txt

Then run:

pip install -r requirements.txt

Make sure you are using the same Python environment that ComfyUI uses. This matters especially if you have multiple Python versions installed.

For portable Windows builds, dependency handling may be different because ComfyUI may use its own bundled Python environment. In that case, use the update or install scripts included with your portable package, or use ComfyUI Manager where possible.

Reduce VRAM Usage

If ComfyUI reconnects when you queue a prompt, especially with FLUX, SDXL, video workflows, ControlNet stacks, upscalers, or large GGUF models, your GPU may be running out of VRAM.

A GitHub issue about connection loss while running FLUX or Z-Image templates describes the UI disconnecting and the command prompt closing when heavier templates were used, while SDXL worked normally. That kind of pattern strongly suggests the problem is related to workload size, model demands, memory, or environment stability rather than the browser alone.

Try reducing image size first. If you are generating at 1024×1024, test at 512×512 or 768×768. Lower the batch size to 1. Disable high-resolution fix, video nodes, or extra upscalers until the basic workflow runs.

You can also close other GPU-heavy apps such as games, video editors, browsers with many tabs, or other AI tools. On systems with limited VRAM, even a few extra gigabytes of usage can cause instability.

Use the Correct Launch Option

ComfyUI has different launch modes depending on your hardware. If you are using an NVIDIA GPU, the normal CUDA launch file is usually the right choice. If your GPU is unsupported or CUDA is not working, testing with CPU mode can help identify whether the issue is GPU-related.

On Windows portable builds, users often start with:

run_nvidia_gpu.bat

If that crashes, try:

run_cpu.bat

CPU mode will be much slower, but if it stays connected, the reconnect issue may be connected to GPU drivers, CUDA, PyTorch, VRAM, or model size.

Check Browser Cache and Extensions

Sometimes the ComfyUI backend is fine, but the browser keeps showing an old frontend state. This can happen after updates.

Try opening ComfyUI in a private or incognito window. You can also test another browser. If ComfyUI works in a different browser, clear the cache for 127.0.0.1:8188.

Browser extensions can also interfere with local web apps. Ad blockers, privacy blockers, script blockers, VPN extensions, and security extensions may block local requests or WebSocket connections.

Disable extensions temporarily and reload ComfyUI. If it works, re-enable them one by one until you find the extension causing the issue.

Check Firewall, VPN, and Antivirus Settings

ComfyUI usually runs locally, but your firewall or antivirus can still interfere with Python, local ports, or browser connections.

Allow Python and ComfyUI through your firewall. If you are using a VPN, turn it off and test again. Some VPNs route or block local traffic in ways that can break tools running on localhost.

If you access ComfyUI from another device on your network, make sure you are using the correct IP address and launch settings. A GitHub WebSocket-related discussion around ComfyUI points toward checking IP and port configuration when troubleshooting connection problems.

For local use, the safest starting point is:

http://127.0.0.1:8188

For another device on the same network, you may need to launch ComfyUI with a listen option and connect through your computer’s local IP.

Avoid Running Broken Workflows Immediately

If ComfyUI reconnects only when loading one workflow, that workflow may be the issue. It may reference missing custom nodes, incompatible model files, invalid paths, or old node versions.

Open ComfyUI with a blank workflow first. Then load the problem workflow. If missing nodes appear in red or the terminal throws errors, fix those before queuing.

A practical example:

You download a workflow from a creator using custom IPAdapter, ControlNet, and video nodes. Your ComfyUI loads the graph, but when you press queue, it reconnects. The issue may not be ComfyUI itself. It may be that one node is missing a model file, one extension is outdated, or your system cannot handle the full workflow.

Start by bypassing heavy sections. Test only the checkpoint loader, prompt, sampler, VAE decode, and save image nodes. Then enable the extra sections one by one.

Reinstall Problem Custom Nodes Cleanly

If updating does not fix a custom node, reinstall it cleanly.

Delete the node folder from:

ComfyUI/custom_nodes

Then reinstall it from its official GitHub repository or through ComfyUI Manager. After reinstalling, install its requirements and restart ComfyUI.

Do not install multiple versions of the same custom node. Duplicate node folders can create conflicts and confusing import errors.

Check Model File Paths

ComfyUI may reconnect or fail if a workflow tries to load a model that is missing, corrupted, or placed in the wrong folder.

Common folders include:

ComfyUI/models/checkpoints
ComfyUI/models/vae
ComfyUI/models/loras
ComfyUI/models/controlnet
ComfyUI/models/clip
ComfyUI/models/unet

If you recently downloaded a model, confirm that it finished downloading fully. A partial .safetensors file can cause loading failures.

Also make sure the model type matches the workflow. For example, an SDXL workflow needs SDXL-compatible models. A FLUX workflow needs the correct FLUX components. Mixing model types can cause errors that look like connection issues.

Update GPU Drivers and PyTorch Carefully

If the reconnect issue started after changing GPU drivers, CUDA, or PyTorch, the problem may be compatibility. Updating everything blindly can make things worse, especially in portable installations.

For NVIDIA users, install a stable recent driver from NVIDIA. Then use the PyTorch version recommended by your ComfyUI installation method. If you are using a portable package, avoid manually replacing core packages unless you know exactly what you are doing.

If ComfyUI worked before a driver or package update, consider rolling back or reinstalling the known working version.

Remote or Cloud ComfyUI Reconnecting Issues

If you run ComfyUI on a remote server, cloud machine, Docker setup, Kubernetes cluster, or reverse proxy, reconnecting can be caused by WebSocket routing.

One GitHub issue describes a hosted ComfyUI setup where WebSocket and HTTP requests could land on different machines because of round-robin load balancing. That kind of deployment can break communication unless routing is configured correctly.

For remote setups, check that your proxy supports WebSockets. Also make sure timeouts are long enough for AI generation tasks. If you run multiple backend replicas, use sticky sessions or ensure the same client stays connected to the same backend instance.

For Nginx-style proxy setups, the important idea is that WebSocket upgrade headers must be passed correctly. Without that, the browser may keep reconnecting even though ComfyUI itself is running.

When Reinstalling Makes Sense

Reinstalling should not be your first move. But it makes sense if your installation has many broken custom nodes, mixed package versions, failed updates, or missing core files.

Before reinstalling, back up:

ComfyUI/models
ComfyUI/output
ComfyUI/input
ComfyUI/user
ComfyUI/custom_nodes

Then install a clean copy of ComfyUI. Test it before adding custom nodes. If the clean version works, add your custom nodes slowly. Do not copy everything back at once, or you may bring the same error back.

The ComfyUI desktop app is also an option for users who prefer a packaged setup. Its GitHub overview says it bundles a stable ComfyUI release, frontend, Manager support, and dependency handling tools.

Common Questions About ComfyUI Reconnecting

Why does ComfyUI reconnect after I press Queue?

This usually means the backend crashed or froze while trying to run the workflow. The most common causes are low VRAM, missing model files, broken custom nodes, or Python dependency errors.

Why does ComfyUI reconnect after updating custom nodes?

Some updates require a full restart. In other cases, a custom node may have updated to a version that needs new dependencies or a newer ComfyUI build. Restart first, then check the terminal for missing packages or import errors.

Can a browser cause ComfyUI reconnecting?

Yes, but it is less common than backend errors. Browser cache, privacy extensions, VPN extensions, and blocked WebSocket connections can cause reconnect loops. Testing in another browser is a quick way to confirm this.

Is “Failed to fetch” the same as reconnecting?

They are related. “Failed to fetch” usually means the browser tried to send or receive data from the ComfyUI server but could not complete the request. That can happen when the backend crashes, restarts, or becomes unreachable.

Should I delete ComfyUI and reinstall it?

Only after you have checked the terminal, updated ComfyUI, disabled custom nodes, and tested a simple workflow. A clean reinstall helps when the environment is badly damaged, but it is not necessary for most reconnecting issues.

Conclusion

Comfyui Keeps Reconnecting can look scary, but it usually has a clear cause. Most of the time, the browser is only showing the symptom while the real issue is in the terminal, custom nodes, memory usage, model paths, or server connection.

Start simple. Restart the backend, check the terminal, test a small workflow, update ComfyUI, and disable custom nodes if needed. If the problem happens during heavy workflows, reduce VRAM usage and confirm your models are installed correctly. If you are using a remote server, check WebSocket and proxy settings.

Once you narrow down whether the issue is caused by a workflow, extension, browser, memory limit, or network setup, the fix becomes much easier. With a careful step-by-step approach, you can stop the reconnect loop and get ComfyUI running smoothly again.