ComfyUI Prompt Manager
Every prompt you run, every image it made, and every negative you tried, in one searchable library inside ComfyUI.
Prompt Manager saves your prompts as you generate, links each image back to the prompt that made it, and gives you one page to browse, tag, compare and reuse all of it. Your library is a local SQLite database on your own machine.
What's new: rebuilt from the ground up
- A new UI. One app, the workspace, replaces the separate dashboard, gallery and metadata pages. The old pages still work if you want them.
- Classic and Nodes 2.0. The nodes work in both ComfyUI renderers. CI starts a real ComfyUI and checks both.
- Tested end to end. Unit tests, browser tests in Chromium, Firefox, WebKit and touch emulation, and tests against a real ComfyUI.
- Mac, Linux and Windows. The Python suite runs on all three, on Python 3.12, 3.13 and 3.14.
See it
All images on this page were generated and recorded through Prompt Manager in a clean demo library.
Your whole output folder, finally browsable
Masonry, grid or list, with search, categories, filters and saved views. Images and videos sit side by side, and selecting one shows its prompts and settings without leaving the page.
Open any image and see how it was made
The viewer shows the positive and negative prompts, tags, seed, steps, CFG, sampler and model next to the image.
Keep your hands on the keyboard
The viewer has shortcuts for moving between files, the slideshow, zoom, rotate, flip, full screen, details, download and delete. Press ? to see them all.
Long libraries stay easy to work in
The selection bar stays pinned while you scroll, so you can keep picking images and act on them from anywhere. Back to top appears once you've scrolled down the page.
One positive prompt, many experiments
The Prompt library groups everything by positive prompt, so repeat runs and negative variants stay together instead of filling your list with near-duplicates.
Watch your negatives branch on a timeline
Each prompt has its own page with a timeline of its negative variants. Click a red dot and the gallery filters to the images that negative made. Small dots mark repeat runs.
Compare negatives term by term
Compare negatives shows two variants side by side and lists the terms one adds, so you can see what actually changed between runs.
Auto-tag your images, and approve what sticks
Tag one image, a selection or your whole output with WD14 or JoyCaption. WD14 suggestions come with confidence scores, and you can review suggestions before anything is saved. Auto-tags belong to the image, so they never change your prompts.
Keep your vocabulary tidy
Tags & categories lists every tag and category with how often it's used. Rename, merge and delete tags, and preview a category change before you apply it.
Group images your own way
Collections are hand-picked sets of images, on top of favorites and saved views. Add from the selection bar and they appear in the sidebar.
Nothing gets guessed
When a file's metadata has more than one possible prompt, Prompt Manager doesn't pick for you. The file waits in Needs review, where you can choose the sampler path, link it to any prompt in your library, or leave it unresolved.
Inspect any file, without importing it
Pick an image or video from your computer to read its prompts, settings and workflow. The file is read in your browser. It isn't uploaded or added to your library.
Send a prompt back to your canvas
Pick an open ComfyUI window and a node, then review the exact before and after. Nothing changes until you apply it.
Deletes you can review
Choose what happens to the files before anything is removed: remove the image from the library and leave the file on disk, or on Linux and macOS also move it to the system trash. On Windows, images can only be removed from the library. A safety backup of the database is on by default, and removed images can be brought back.
Maintenance in one place
System holds settings, image sources, auto-tag models, integrations, maintenance, backups and logs. Risky operations show a review first, and long ones run as jobs you can follow in Activity.
Works on your phone too
The workspace adapts to narrow screens, and touch is part of the browser test suite.
The nodes
Six nodes, under ComfyAssets > Prompts in the Add Node menu. Pick the one that suits your workflow; they all share one library. Every screenshot below is a small real workflow that ran.
Prompt Manager (Classic)
The drop-in replacement for CLIP Text Encode: same CLIP input, same CONDITIONING output, with the prompt saved as it runs. It also outputs the text as a STRING.
The same workflow in Nodes 2.0:
Prompt Manager (Focused)
An editor-first layout. Composition and metadata fold away, and Browse library lets you search, preview and use a saved prompt right on the node.
Prompt Manager (Split)
The same editor with the library always visible. Wide nodes put them side by side; narrow nodes stack them.
Prompt Manager (+/-)
Positive and negative in one node, saved as a pair. This is the node that feeds the negative-variant timeline. It outputs CONDITIONING and STRING for both sides.
The same workflow in Nodes 2.0:
Prompt Manager Text
Saves the prompt and outputs a STRING instead of CONDITIONING, for workflows that send text to another node. Prepend and append inputs let you build the final text.
Prompt Search List
Searches your library by text, category, tags and minimum rating, then outputs the matches as a list so ComfyUI runs each one. This workflow found three prompts tagged mushroom and made three images.
Install
Requirements: ComfyUI and Python 3.12 or newer. watchdog and Pillow are the only dependencies of their own; the rest comes with ComfyUI.
Install from the Comfy Registry or ComfyUI-Manager (search for Prompt Manager), or by hand:
cd ComfyUI/custom_nodes/
git clone https://github.com/ComfyAssets/ComfyUI_PromptManager
pip install -r ComfyUI_PromptManager/requirements.txt
Restart ComfyUI. Click PM in the top bar, or Web UI on a node, or open http://localhost:8188/prompt_manager/workspace.
Optional extras. Nothing here is installed for you:
| To use | You need |
|---|---|
| WD14 auto-tagging (about 400 MB per model) | onnxruntime, or onnxruntime-gpu for CUDA |
| GGUF JoyCaption (about 4.5 GB) | llama-cpp-python |
| HuggingFace JoyCaption (about 16 GB) | transformers, accelerate, bitsandbytes |
| Prompts embedded in videos | ffprobe on PATH |
| LoRA Manager import | LoRA Manager installed; turn it on in System > Integrations |
Download or unload tagging models in System > Auto-tagging.
Where your data lives
| What | Where |
|---|---|
Library (prompts.db), config.json, database backups, logs |
ComfyUI/user/__promptmanager/ on new installs |
| Thumbnails | output/thumbnails |
| Your images | Left where they are. Prompt Manager reads your output folder and any sources you add. |
Set PROMPT_MANAGER_DATA_DIR to an absolute folder to keep the library somewhere else. config.json can hold your CivitAI key, so treat it as private.
Upgrading from an earlier version
- Python 3.12 or newer is now required.
- Your library stays where it is until you choose. The workspace offers Move library, which makes a verified copy in
user/__promptmanager/and renames the old files instead of deleting them, or Keep it here. - Older images: a one-time notice offers a metadata scan that links them to prompts and negatives. Scans don't count as new runs.
- Going back isn't supported. Workflows are now stored once, and an older version shows new records without their workflow or prompt graph. Back up first in System > Backup & export. See the old README's changelog for the manual way back.
The old pages still work
They stay at their URLs, and System's Overview section links to them:
| Page | URL |
|---|---|
| Dashboard | /prompt_manager/admin |
| Tags | /prompt_manager/admin#/tags |
| Gallery | /prompt_manager/gallery |
| Metadata viewer | /prompt_manager/gallery.html |
The documentation for those pages is in README-legacy.md.
Known limits
- Trash is Linux and macOS only. Windows has no Recycle Bin option, because the bin can delete large files permanently without warning. On Windows you can only remove an image from the library, and the file stays on disk. Where a move to the trash isn't possible on Linux or macOS, the dialog lists why for each file.
- Show in folder opens the file manager on the machine running ComfyUI, not necessarily the one with your browser.
- Returned images get their prompt links back, but not their tags, categories, collections or favorites.
- Inspect a file doesn't add the file to your library.
- Leave unresolved is permanent. A file you leave unresolved in Needs review isn't offered again unless its bytes change.
- Pair prompts saved by earlier versions stay Uncategorized. You can set the category with Edit metadata.
How it's tested
| Layer | What runs |
|---|---|
| Python | Over 200 test files on Linux and Windows with Python 3.12 to 3.14, with an 85% coverage gate |
| Frontend | About 100 unit test files, with coverage floors per file |
| Browser | Over 60 Playwright spec files in Chromium, Firefox, WebKit and touch emulation |
| Real ComfyUI | A pinned CPU ComfyUI, started in Classic and Nodes 2.0, running real workflows |
Linux runs on every pull request and in the merge queue. Windows runs on every push to main and nightly. macOS is not tested; it should work, but support is best-effort. To run the tests yourself, see TESTING.md.
License
MIT. See LICENSE.