In short
For a recurring character you need four custom node packs, not forty: ComfyUI-Manager to install the rest, Impact Pack for FaceDetailer, rgthree-comfy for a Power Lora Loader, and VideoHelperSuite for reels. IPAdapter plus has been maintenance-only since 14 April 2025. Install one pack at a time and restart, as the ComfyUI docs advise.
Custom nodes are where a character workflow quietly stops working: you install one pack for a face fix, a second for video, and a week later the graph shows red boxes. Every pack you add is code that has to agree with ComfyUI and with every other pack on the same Python packages. The ComfyUI docs call dependency conflicts a common issue. So the question for a recurring face is not which packs are popular, it is which few you cannot do without.
This list ranks seven packs for that job. We did not benchmark them. Licenses, features and maintenance notices come from each repository's README as we read it on November 4, 2026, and the problems come from a public ComfyUI discussion and the official docs, all linked in the sources. We left out star counts, because a count says what was popular, not what is safe to install this week. If a README shows no release or date, we say nothing about versions.
How we ranked them
- Does it fix a job a character feed has? A consistent face, a fixed hand or face on a wide shot, several LoRAs at once, a reel from stills. A pack that does none of these is a convenience, not a need.
- Maintenance status. Whether the README says the project is active, in maintenance mode, or says nothing. A pack that stops following ComfyUI is a future red box.
- License. The code license of the pack, read from the repository. It says nothing about the models you run through it; see our note on model licenses.
- Install cost. Extra steps and extra Python packages. The fewer a pack drags in, the less it can conflict.
- Is it replaceable? If a built-in or smaller pack does the same, the pack drops down the list.
The ranking
- 1
ComfyUI-Manager
Install it first and use it for everything else. It installs, removes, disables and enables custom nodes, manages models and snapshots, and detects missing nodes, per its README and the official docs. It is what turns a red box into one click, and what Impact Pack's own README recommends for installing.
Best for: Anyone who opens other people's workflows, which is every character creator
- GPL-3.0 license
- Detects missing nodes in a workflow you load, which is the failure our guide on [missing nodes](/guides/comfyui-missing-nodes) covers
- Snapshots let you go back after an update breaks something
- Documented on docs.comfy.org
- It is the tool that makes one-click installs easy, which is also how people end up with thirty packs
- The README we read shows no releases section, so we cannot say how often it ships versions
- 2
ComfyUI-Impact-Pack
The face-fix pack. Its README describes FaceDetailer as a node that "easily detects faces and improves them", and a FaceDetailer (pipe) variant for multipass. For a feed where the face has to match on a wide shot, this is the one pack that earns its weight.
Best for: Re-rendering a small face in a full-body or wide shot so it matches the close-ups
- GPL-3.0 license
- FaceDetailer plus SAM-based detector nodes in one pack
- README notices are dated by version, so you can see what changed (V8.24 needs ComfyUI 0.3.63 or higher)
- Recommends installing through ComfyUI-Manager
- The UltralyticsDetectorProvider node is not part of it; the README says to install ComfyUI-Impact-Subpack separately, and since V8.0 that is no longer automatic
- Since V7.6 automatic installation is no longer supported, so a manual clone needs its requirements installed in ComfyUI's own Python environment
- As we read its dependency list it limits numpy to below 2, the kind of pin that collides with other packs
- 3
rgthree-comfy
Quality of life that matters for characters. The Power Lora Loader loads several LoRAs at once in one condensed node with a toggle for each, which is how a character LoRA, a style LoRA and a lighting LoRA stay manageable. Image Comparer and the Fast Muter make A/B testing a new checkpoint on the same face quick.
Best for: Stacking a character LoRA with others and testing changes on the same seed
- MIT license, the most permissive on this list with WAS Node Suite
- Power Lora Loader, Image Comparer, Fast Muter and Bypasser, Seed control and a Link Fixer for broken workflow links
- Does not need a detector model or a large install
- Convenience, not capability: every function has a plainer core equivalent, just with more nodes
- The README is a feature list; it gives no compatibility table for new model families
- 4
ComfyUI-VideoHelperSuite
The video plumbing pack. Load Video turns a clip into frames with options for frame rate, size and frame limit, Video Combine writes frames back out with audio and a chosen format, and you can define new ffmpeg output types with a JSON file. If you cut reels from a character, this is how the frames get in and out.
Best for: Image-to-video reels, frame extraction and re-encoding with audio
- GPL-3.0 license
- Load Video, Load Image Sequence, Load Audio and Video Combine
- Batch helpers: split, merge, select every Nth, duplicate
- Custom output formats through JSON files in a video_formats folder
- Useless for a stills-only feed
- Advanced Previews needs ffmpeg, which is another thing to install
- ComfyUI 0.44.0, released today, lists ProRes output in core, so check whether core already covers your export before adding it
- 5
ComfyUI-KJNodes
A grab bag. The README describes it in one line, "various custom nodes for ComfyUI", and many popular public workflows lean on it for small utilities. Install it when a workflow you want to run reports it missing, not before.
Best for: Running someone else's workflow that lists it as a requirement
- GPL-3.0 license
- The most recent dated changelog entry we could read is March 19, 2026, so it is not abandoned
- No stated focus, so no reason to install it blind
- A broad pack means a broad dependency surface
- 6
WAS Node Suite
A very large toolbox: the README counts 678 nodes covering image processing, masking, text, logic, latents and more. It is permissively licensed and was the default recommendation of older roundups. For a character feed it is mostly more than you need, and the size is the cost.
Best for: Text and string handling, masks and filters when no smaller pack has the node
- MIT license
- One install covers many small jobs
- 678 nodes to load for the two or three you use
- Any pack of this size is a bigger target for conflicts; we found no maintenance notice on the page either way
- 7
ComfyUI IPAdapter plus
The classic reference-image tool, "a 1-image lora" in its own words, and the reason for most old consistent-face graphs. But its author wrote on 2025.04.14 that the repository is in "maintenance only" mode and that they do not plan consistent work on it. Do not build a new feed on it. For a face you keep, train a character LoRA or use a model's own reference editing.
Best for: Keeping an existing SDXL-era workflow running
- GPL-3.0 license
- Many example workflows in the repository
- Maintenance-only since 2025.04.14, per the author's notice
- Reference-image conditioning is a weaker answer to a recurring face than a trained LoRA, see [train a character LoRA](/guides/train-a-character-lora)


The cheat-sheet
| Pack | License | Does for a character feed | Status note |
|---|---|---|---|
| ComfyUI-Manager | GPL-3.0 | Installs packs, finds missing nodes, snapshots | Documented on docs.comfy.org |
| Impact Pack | GPL-3.0 | FaceDetailer, SAM detectors | Ultralytics detector needs the Subpack; V8.24 needs ComfyUI 0.3.63+ |
| rgthree-comfy | MIT | Power Lora Loader, Image Comparer, muters | No notice found |
| VideoHelperSuite | GPL-3.0 | Load Video, Video Combine, batch helpers | No notice found |
| KJNodes | GPL-3.0 | General utilities | Last dated changelog entry March 19, 2026 |
| WAS Node Suite | MIT | 678 image, mask, text and logic nodes | No notice found |
| IPAdapter plus | GPL-3.0 | Reference-image conditioning | Maintenance only since 2025.04.14 |
Our verdict: four packs, installed one at a time
The kit for a character feed is Manager, Impact Pack, rgthree-comfy and, if you make reels, VideoHelperSuite. Add KJNodes or WAS Node Suite only when a workflow you want to run asks for them. Leave IPAdapter plus for old graphs. Four packs cover face repair, LoRA stacking, A/B testing and video in and out, and four is few enough to find the culprit when something turns red.
The order of work matters more than the list. The ComfyUI docs say to try installing plugins one by one, restarting ComfyUI after each installation to observe if dependency conflicts occur. We would add one habit: before a big model change such as the move to a newer base, see which still-image model keeps the same face, take a Manager snapshot first. Then a broken update is a restore, not a rebuild.
What creators are saying
A ComfyUI update broke one custom node. Rolling ComfyUI back fixed that node and broke others, which is how the thread's author came to ask for custom nodes to be treated as Python dependencies.
ComplaintComfyUI discussion on custom nodes as Python dependencies, November 2023
Some nodes depend on specified versions and have conflicts, so running different workflows in one environment is hard.
ComplaintSame discussion, August 2024
A collaborator on the Manager explained that strict dependency management can make resolution impossible depending on install order, and that the Manager therefore installs packages individually.
Custom node requirements are not standardized, and numpy and timm in particular are prone to version conflicts.
ComplaintSame discussion, December 2024
What breaks, and what to do first
- 1
Snapshot before you change anything
Manager lists snapshot management among its features. Take one before updating ComfyUI or adding a pack, so a bad update can be reversed.
- 2
Install one pack, restart, open a known workflow
This is the docs' own advice. If nodes vanish from the library or an error pop-up appears, the last pack you added is the suspect.
- 3
For Impact Pack, install the Subpack only if you need Ultralytics detectors
The README says the Ultralytics detector node lives in a separate pack since V8.0. If you do not use YOLO face detectors, skip it and keep one fewer set of packages.
- 4
Check the pack's notice list against your ComfyUI version
Impact Pack's V8.24 note says it needs ComfyUI 0.3.63 or higher. Notices like this are where an old install and a new pack disagree.
- 5
Remove what you did not use
A workflow that asked for KJNodes once does not need it forever. Disable packs you have not opened in a month.

Where Musetta fits
Musetta is our studio. It runs on your own ComfyUI, installs the nodes and models a workflow is missing, and keeps each character's reference photos, so adding a pack is a step it handles and not a Python session. It is not a replacement for knowing what you installed; the table above is worth keeping either way. See how Musetta compares with working in ComfyUI directly and the features.
FAQ
Short answers are below in the structured FAQ. For a model to run through these nodes, start with MOSS Image in ComfyUI and the guide to a consistent AI character in ComfyUI.
Questions
- Which ComfyUI custom nodes do I need for a consistent character?
- Four are enough for most feeds: ComfyUI-Manager to install and snapshot, Impact Pack for FaceDetailer, rgthree-comfy for its Power Lora Loader, and VideoHelperSuite if you make video. Add others only when a workflow reports them missing.
- Is ComfyUI IPAdapter plus still maintained?
- Its author wrote on 2025.04.14 that the repository is in maintenance only mode, that crucial updates or PRs might still be merged, and that no consistent work is planned. We would not start a new character feed on it.
- Why does Impact Pack not show the Ultralytics detector node?
- The README says UltralyticsDetectorProvider is not part of the pack and needs ComfyUI-Impact-Subpack, which since V8.0 is no longer installed automatically.
- How do I avoid custom node dependency conflicts?
- The ComfyUI docs advise installing plugins one at a time and restarting ComfyUI after each, to see whether a conflict appears. Take a Manager snapshot first and keep the number of packs small.
- Are these custom node packs free for commercial use?
- Their code licenses, read from the repositories, are GPL-3.0 for Manager, Impact Pack, VideoHelperSuite, KJNodes and IPAdapter plus, and MIT for rgthree-comfy and WAS Node Suite. That is about the code. The base model you run has its own license, which governs your images; check it separately.
Sources
- 01ComfyUI-Manager · GitHub, 2026-11-04
- 02ComfyUI-Manager overview · ComfyUI docs, 2026-11-04
- 03ComfyUI-Impact-Pack README · GitHub, 2026-11-04
- 04rgthree-comfy README · GitHub, 2026-11-04
- 05ComfyUI-VideoHelperSuite README · GitHub, 2026-11-04
- 06ComfyUI-KJNodes README · GitHub, 2026-11-04
- 07WAS Node Suite README · GitHub, 2026-11-04
- 08ComfyUI IPAdapter plus README · GitHub, 2025-04-14
- 09Dependencies, custom nodes and conflicts · ComfyUI docs, 2026-11-04
- 10Dependency Resolution and Custom Node Standards · Comfy blog, 2025-06-07
- 11Custom Nodes as Python Dependencies, discussion #1959 · GitHub, 2024-12-04
- 12ComfyUI 0.44.0 · freedom.tech, 2026-11-04
Researched and written by the Musetta editorial desk with AI assistance, from the sources above. Numbers come from those sources; nothing is benchmarked by us unless the article says so. Spotted an error? Tell us on Discord and we fix it.