WeiLin-Comfyui-Tools turns ComfyUI prompt tags into a managed library with previews

The unit of work is a tag, and tags live in groups you can sort and share

ComfyUI is a node graph for image generation, and a prompt is usually a wall of comma-separated tokens. WeiLin-Comfyui-Tools is a custom node pack built around turning that wall into a set of named, grouped, searchable items called tags. The readme describes a tag manager interface where every tag card carries a 120 by 120 preview area, and the surrounding feature set has grown over a long series of public beta releases. Groups can be renamed, reordered and resized, the search matching is described as more precise, and a random tag feature lets you define templates and generate a random selection in one action, with each node holding its own customizable template. Tags also carry weights, including negative ones, and there is a per-node switch for random tag loading. Two features from the 0.0.62 release are worth singling out because they change how the library is used rather than how it looks: a share function for tag sub-directories, and a one-click clear that empties the whole collection. The repository keeps a `tag_labels.json` file at its root, which suggests the label structure is data the plugin ships rather than something it derives from your graph.

Preview images are generated by ComfyUI's own queue, not by a second renderer

The preview feature, added in the 0.0.77 through 0.0.79 releases and credited to pull request 75, is the part of this project with real engineering in it. The readme says the generation uses ComfyUI's internal queue API for text-to-image, which means the plugin does not carry a diffusion model or a sampler of its own. It submits a job into the same queue that your own workflows use, and the results come back as thumbnails. Because the work goes through the queue, the plugin inherits the properties of a queue: tasks execute in order in a background async queue, a progress bar shows completed against total for batch runs, there is a cancel button that stops all queued and running tasks at once, and if you refresh the page the polling state resumes and picks up where it left off. The parameter dialog is where the model comes from. The readme says the model selection scans the ComfyUI checkpoints directory automatically, and the rest of the dialog covers size presets such as 512x512, 768x512 and 512x768, a KSampler choice with a scheduler, adjustable steps, CFG and seed, and editable positive and negative prompts. Parameters are saved locally, so you set them once. Hovering a thumbnail for 200 milliseconds shows the full-size image in a floating preview.

The translation layer moved from keyword APIs to AI and back again

The changelog is the clearest description of this project's history, and the translation subsystem is where it gets interesting. In 0.0.66 the translation feature was rewritten to use AI, calling an API on SiliconFlow, with the note that requests for other AI platforms should be raised in the group so they could be added later. Then in 0.0.70, 0.0.71 and 0.0.72 the API translation mode returned, and the interface was rewritten so that it now supports the Alibaba translation interface, the Bing translation interface and the Youdao translation interface, with a warning that the Youdao interface has request limits and is not recommended. All three are described as free, with the note that paid interfaces would be added later. The SiliconFlow and OpenAI problems were fixed in the same releases by reverting to the original mode. Reading those three entries together tells you the shape of the project: a personal tool where the author experiments with a different translation backend, finds the practical problems, and rolls back to what works. Version 0.0.72 also merged pull request 57, and 0.0.69 merged pull requests 50 and 52, so community contributions have been landing in this area.

LoRA management is a separate surface with a Civitai-shaped data path

Alongside the tag manager there is a LoRA manager, and the changelog shows it being fleshed out over several releases. From 0.0.48 onwards the manager got a floating card that shows basic information when you hover a LoRA entry, a detail panel where you can define your own fields, and a fix for the second-level directory not loading. Version 0.0.59 added the ability to load a LoRA as if it were a tag, and it says the plugin recognises it in the form of a wlr prefixed tag carrying a name, a model weight and a text weight, and that this only takes effect in the all-in-one prompt editor. Version 0.0.60 added a Danbooru manager for handling Danbooru data more conveniently. The Civitai connection shows up in the 0.0.52 entry, where fetching LoRA data from that site also fetches the cover image. The repository layout backs this up, with separate `lora_userdatas/` and `translate_userdatas/` directories alongside `user_data/` and `random_tag/`, which suggests the plugin stores its own per-category state on disk rather than only in the ComfyUI graph.

Installing it means the Comfy Registry path, and the docs push you to a chat group

The packaging tells you how it is meant to arrive. The pyproject.toml names the package weilin-comfyui-tools, version 0.0.80, with a licence that points at a LICENSE file, and it carries a tool.comfy block with a publisher id, a display name and an icon, with a comment in the file noting it is used by the Comfy Registry. The runtime dependencies are minimal, and the requirements file lists exactly two:

uuid7
aiosqlite

That is the whole dependency surface, one identifier generator and one asynchronous SQLite driver, which is a good sign for a plugin that runs inside someone else's Python process. The Python side is in `src/`, with `app/`, `js_node/` and `random_tag/` directories, and the project targets py310 with a line length of 88 for both Ruff and Black. The readme itself does not contain a command-line install procedure. Instead, the install and usage questions are directed to an official QQ group, and the readme says the plugin is still in internal testing while tending towards a stable state, with the group number given as 905656521 after an older one was closed. The project also points at two companion repositories: a separate panel project for migrating old plugin prompts and for offline tag editing, and a prompt repository that holds the cloud tag and Danbooru database.

The release naming and the author's own statement set expectations

Two facts in the readme matter for anyone depending on this. The first is the author's statement about update frequency. It says that because personal time is limited the plugin will not be updated often, that updates may happen once in a while when there is free time, that each update tries to satisfy the requests raised, and that there is generally no large bug, so the update rate is low, described as two to five times a month. That is a candid statement and it is a more useful signal than a release cadence table. The second is the version scheme. The recent releases are tagged with a pre suffix, at v0.0.72pre, v0.0.71pre and v0.0.70pre, and the readme calls each of these a public beta release, with the newest entry for 0.0.80 dated 2026-09-25. A plugin in public beta with pre-release tags and a self-declared slow update rate is a tool for a personal machine, not something to build a production pipeline on. The licence recorded in the repository metadata is GPL-2.0, and the readme carries an author statement of copyright, so a project that redistributes it as part of another tool has a copyleft obligation to consider.

Where this plugin is the wrong choice

Three limitations are visible in the readme rather than inferred. The first is the dependence on ComfyUI internals. The preview generator calls ComfyUI's internal queue API, and the changelog contains a long series of fixes for node loading bugs, including a fix for the case where a node is not connected or not used, where the readme warns you to turn off the random prompt switch on a node you are not using. Working against internal APIs means an upstream ComfyUI release can change behaviour underneath a plugin, and the fix arrives when the author gets to it rather than when upstream ships. The second is that the feature set is broad and shallow in places, with a large amount of it being tag and LoRA organisation, which is genuinely useful but overlaps with what the prompt editor in ComfyUI can already do for a single workflow. The third is the support model. Usage questions, the panel for migrating old prompts, and the roadmap for additional AI platforms all route through a chat group rather than an issue tracker with triage, even though issues and requests are invited for the plugin itself. A team that needs a support contract cannot get one from this repository. The alternative to weigh against is writing a small custom node that loads a prompt list from a file, which gives up the manager, the previews and the translation but replaces the plugin dependency with a directory you own.

Editorial conclusion

WeiLin-Comfyui-Tools is worth a look for a ComfyUI user whose prompt workflow has turned into a filing problem, since the tag manager, the group structure, the random tag templates and the per-tag preview images address exactly that, and the preview feature runs generation through ComfyUI's own queue rather than a second process. It is a poor fit if you need offline or headless operation, because the translation modes call external services including the Alibaba, Bing and Youdao interfaces and an AI platform, and it is a poor fit for a shared production deployment, because the readme describes a personal tool with chat-group support rather than a maintained service contract. Before adopting it, read the licence file, since pyproject points the package licence at a LICENSE file and the repository lists GPL-2.0, check which ComfyUI version you run against the node bugs the changelog mentions, and note that the last push to the repository was on 2026-09-25 with the version in pyproject.toml at 0.0.80.

Frequently asked questions

Official sources

1. License: GPL-2.0
2. Project website
3. README
4. Releases
5. weilin9999/WeiLin-Comfyui-Tools on GitHub