ComfyUI Video Nodes Not Showing: Find and Fix the Missing Step

AI Overview

Why are my ComfyUI video nodes not showing?

The node may belong to an uninstalled package, fail during Python import, or require a newer ComfyUI version.

Does installing a video model add its nodes?

No. Model weights and node code are separate. Install the correct node package and its dependencies, then restart.

Why is VideoHelperSuite installed but still missing?

Check the startup log and active Python environment. A package folder can exist even when its nodes fail to load.

Should I reinstall ComfyUI?

First identify the missing node type and test a small workflow. Repair the affected package before rebuilding everything.

Start With the Symptom, Not a Reinstall

ComfyUI video nodes not showing is usually a setup problem that happens before generation. A downloaded graph opens with unknown boxes, a tutorial's node name does not appear in search, or a video loader seems unavailable. Those symptoms are different from a workflow that runs successfully and then fails to display its finished movie.

Before changing anything, save a copy of the workflow and record the missing node names exactly. Keep the current startup log. Write down whether you use Desktop, Windows Portable, a manual Python environment, or a hosted installation. A command that repairs one environment may change an unrelated Python installation on another machine.

Use this decision table to choose the first investigation. It is a practical diagnosis aid, not a claim that one symptom proves a single cause.

If the queue completes before the problem appears, use the separate ComfyUI freeze recovery guide. This article focuses on restoring missing functionality before you queue the expensive production graph.

Identify the Exact Node and Its Owner

A friendly label is not always a unique identifier. Workflow authors can rename boxes, and different packages can offer similarly named video loaders. Inspect the original workflow JSON without editing it. Standard graph exports describe node types; API exports commonly identify each node with a class_type field. Record that identifier alongside the visible label.

Then compare the workflow's requirements with the repository or official template from which it came. Separate core ComfyUI nodes, custom packages, and hosted API nodes. Updating a custom package cannot add a newer core feature to an older backend. Likewise, downloading a large video checkpoint does not install the Python class that understands it.

For VideoHelperSuite, look for the relevant VHS loader or combine operation in the package's own documentation. For a Wan wrapper workflow, verify the wrapper's requirements rather than substituting a native Wan node with a similar name. Similar input labels do not establish equivalent socket types, model handling, or behavior.

Keep a small ownership record: workflow filename, missing type, owning package, installed revision, and required revision if the author specifies one. If the author provides no dependency list, report the exact unknown type instead of searching only for “video node.” That information makes a support request actionable and avoids installing unrelated packs.

Install in the Environment That Actually Runs ComfyUI

Check the launcher's location before installing anything. Installing into one environment while opening another leaves the symptom unchanged.

Use ComfyUI Manager where available to locate the identified package. Confirm its publisher and repository against the workflow requirements, install it, inspect the installation result, and restart the backend. If manual installation is required, follow that repository's instructions for placing the code under the active installation's custom-node directory and installing its dependencies.

For manual Python setups, use the interpreter that launches ComfyUI. The following checks are diagnostic examples to run within that environment, not commands to paste into an arbitrary system terminal:

python -c "import sys; print(sys.executable)"
python -m pip --version
python -m pip check

The first two outputs should point to the intended environment. The third reports dependency inconsistencies; it does not automatically fix them. On Windows Portable, the bundled interpreter is typically under python_embeded. Use the actual bundled executable when following package installation instructions. Desktop users should follow Desktop's supported environment-management path.

After restarting, search for the exact type in a blank canvas. Reopen the original workflow only after the node appears. Record two separate outcomes: installation succeeded and node registration succeeded. Treating those as one event is what makes an installed-but-missing problem so confusing.

Keep unrelated packages unchanged during this test. Updating every dependency at once makes it harder to identify which change restored the node or broke another workflow.

Fix Import Errors and Version Mismatches

Read the startup traceback associated with the missing package. A missing module, an incompatible API import, and a failed compiled extension need different repairs. Preserve the first useful exception rather than reporting only the final “import failed” summary.

For a missing dependency, verify that the package's documented requirements were installed into the active interpreter. For a renamed API or unavailable function, compare the package revision with the ComfyUI revision it supports. For compiled dependencies, check the package's supported Python, PyTorch, and platform combination before selecting a wheel.

A newer tutorial can also target features that have not reached your stable installation. Check the official release and update instructions for your installation type. Do not assume that refreshing the browser updates the backend, or that updating backend source also refreshes every separately installed dependency and frontend component.

Make one controlled change, restart, and compare the log. If nodes disappear after an update, retain the before-and-after versions. Compare against a known working environment instead of repeatedly modifying the same setup without notes.

When interface extensions are implicated, ComfyUI's official troubleshooting approach supports disabling extensions and narrowing the cause by groups. Remember that disabling all custom nodes intentionally removes their functionality: use a core-only control workflow for that test. Failure of the original custom-node graph in that mode is expected and does not identify the culprit.

Once the relevant package imports cleanly, stop changing dependencies. Move on to functional verification instead of continuing to tune an environment that now satisfies the missing-node requirement.

Separate Missing Models From Missing Video Output

If a loader exists but its dropdown is empty, node installation has progressed further than the original symptom suggests. Check the model category expected by that specific loader, the configured model search paths, and whether the intended file finished downloading. Follow the workflow's model manifest; do not put every video-related file into the checkpoints directory.

A file appearing in a dropdown does not prove architectural compatibility. A VAE, text encoder, and diffusion model must match the workflow's requirements. Renaming an incompatible file cannot convert it. For decoder-specific failures, the VAE loading checklist helps keep file discovery separate from successful model loading.

If VideoHelperSuite nodes are visible but a preview is missing, inspect the saved file before reinstalling. Its loader converts video into frames, while Video Combine assembles frames into a video and can include an audio input. Frame rate affects playback duration, so preserve it deliberately when making a round trip.

Preview behavior is another layer. VideoHelperSuite documents optional advanced previews that use FFmpeg and can introduce a conversion delay. A blank or delayed preview is therefore different from an absent node class. Check whether the file opens independently, whether the selected format is supported, and whether logs identify an encoder error.

For audio-sensitive tests, use the crop video with audio walkthrough to keep the source, frame timing, and sound path explicit. Do not judge a repair from a single still thumbnail.

Existing AI-generated illustration: use a visually simple subject for inspection. This is not a ComfyUI test result.

Prove the Repair With a Small Video Test

Use a short clip whose contents you already understand. For a video utility package, start with loading and saving existing footage before involving a generative model. This separates registration and basic media handling from GPU memory, model compatibility, and sampler behavior.

Run this acceptance sequence and record the actual result at each gate:

1. Restart: the relevant package loads without its previous import error.
2. Search: the required node can be added to a blank graph.
3. Import: the original workflow recognizes its required types.
4. Input: a short source loads with the expected dimensions and frames.
5. Output: the saved movie plays through with the intended timing and audio.
6. Repeat: a second run succeeds after reopening the saved test workflow.

Existing motion and audio example for a video round-trip check

This existing Seedance media example illustrates what to inspect in moving footage. It is not evidence that we reproduced or repaired your ComfyUI environment.

For example, if loading succeeds but exporting fails, keep the working loader and investigate the output stage. If both work in the utility test but the generative workflow fails, return to model and sampling diagnostics. The smallest passing test creates a useful boundary between repaired functionality and a separate problem.

Save the successful JSON with a brief environment manifest. Include ComfyUI, package revisions, Python, and the output settings. Once individual clips work, the batch video workflow guide can help you scale the process without turning one failed item into a repeated full batch.

Keep Production Moving While You Repair

Maintain a repair track and a delivery track when a deadline matters. The repair track owns the exact node, startup error, environment, and minimal reproduction. The delivery track owns the creative brief, source references, approved shots, and outstanding edits. This prevents infrastructure troubleshooting from erasing creative decisions that were already settled.

Local ComfyUI remains useful when custom models and exact graph control are part of the job. A hosted workflow is useful when your immediate requirement is an approved clip and nobody needs to reproduce the original graph. The outputs and controls may differ, so compare them against the same brief rather than assuming they are interchangeable.

Seedance Agent provides a place to organize references, plan shots, review results, and request focused revisions. It is an alternative production route, not a repair tool for Python imports. Carry over the required action, aspect ratio, reference identity, duration, and acceptance criteria; reassess the generated result before inserting it into the edit.

Conclusion

Fix ComfyUI video nodes not showing by identifying the exact node owner, checking the active environment, reading import errors, and verifying a short video round trip. Keep missing weights and blank previews separate from missing node code, then save the working setup before restoring a large graph. If your immediate priority is finishing the shot, create a video with Seedance while you isolate the local issue.