Extensions/comfyui-gprompts

ComfyUI Extension

comfyui-gprompts

G-Prompts - capture the text prompt that produced a result, review your library, and feed prompts back into workflows, right inside ComfyUI.

By AI4VFX·Created 2 months ago·Updated about a month ago· 3

AI4VFX/comfyui-gprompts

Nodes—

On cloudLocal install

Stars3

Updatedabout a month ago

Readme

<img src="assets/G-Prompts_Promo.png" alt="G-Prompts" width="820"> <h1 align="center">G-Prompts</h1> Capture the prompt that produced a result, review your library, and feed prompts back into any workflow — right inside ComfyUI. <img alt="ComfyUI custom node" src="https://img.shields.io/badge/ComfyUI-custom%20node-1f6feb"> <img alt="version" src="https://img.shields.io/badge/version-1.1.0-2ea043"> <img alt="license" src="https://img.shields.io/badge/license-MIT-8957e5">

G-Prompts is a native ComfyUI sidebar for building a personal, reusable prompt library. Hit one button and it traces your active workflow to grab the exact positive/negative prompt that made your last image, video, or audio — bundled with a thumbnail of the result and a copy of the workflow itself. Later, browse your library, edit prompts in place, and drop them back onto the canvas.

No graph nodes are added; it's UI + a handful of local server routes. Your library is just files on disk in a folder you choose — one card is a small set of sidecar files (.txt prompt, .json metadata, a thumbnail, and the workflow), so it stays readable, portable, and safe to back up or edit by hand.

▶ Walkthrough

<a href="https://github.com/AI4VFX/comfyui-gprompts/releases/download/v1.0.0/G-Prompts_Walkthrough.mp4"><img src="assets/G-Prompts_Walkthrough.png" alt="Watch the walkthrough" width="640"></a> ▶ Download and watch offline (~4 min walkthrough · 56 MB)

Prefer short clips? Jump to the feature demos below — fifteen 10-second videos, one per feature.

Install

ComfyUI Manager (recommended) — search for G-Prompts in the Manager and install, then restart ComfyUI.

Manual

cd ComfyUI/custom_nodes
git clone https://github.com/AI4VFX/comfyui-gprompts.git

Restart ComfyUI. No Python dependencies — G-Prompts runs on the standard library alone.

ffmpeg is optional — it's only used to make compact thumbnails/posters for video results. Put ffmpeg on your PATH if you generate video; images and audio don't need it.
First launch: open the Prompts tab and pick a folder to store your library in.

What's new in 1.1.0

Non-destructive delete — "Delete" now moves cards into a Trash folder inside your library (mirrored hierarchy, one-click Restore, size badge). G-Prompts ships with zero deletion code: nothing it does can permanently destroy a file. You empty the Trash yourself, in your file manager, when you decide.
Capture input assets — optionally zip the input images/video/audio your workflow references beside the card (.assets.zip), and include them in Packs. A captured card can now carry everything needed to reproduce the result.
Remember selected folder — reopen the panel in the folder you last used.
Hide / unhide a card's custom thumbnail, with an "!" badge when hidden.
Per-card default thumbnail — choose which built-in placeholder (Image / Video / Audio / Other) a card shows.
Card sorting — toolbar button cycles ↓ Newest → ↑ Oldest → A→Z → Z→A.
⟳ Refresh toolbar button — rescan the library so files you dropped in by hand get picked up.
Pack copies the path — packing now puts the zip's full path on your clipboard and shows a toast (duration configurable in Settings).
Details pane on by default for fresh installs; new "A" (assets) badge on cards.

Getting started

Open the Prompts tab in the ComfyUI left sidebar.

On first run you'll be asked to choose a library folder — everything G-Prompts saves lives there. That's it; you're ready to capture.

Features

⚡ Capture

Click Capture (or press the shortcut, Alt+Q by default) and G-Prompts traces your current workflow to find the prompt that feeds generation:

Sampler workflows → reads the sampler's positive/negative conditioning and resolves it back to the original prompt text (skipping enhancement sub-graphs and parameter strings).
No-sampler workflows (LTX, TTS, custom, VLLM) → walks back from the output/save nodes to the prompts actually wired into the result.
Selection → capture exactly the text node(s) you've selected, split into positive/negative by wiring, node title, or sampler branch.

Each capture becomes one card holding the positive and negative together, plus a thumbnail of the latest result and a copy of the workflow. Captured cards keep the original prompt (stable and reproducible), not a runtime-enhanced variant. With Capture input assets enabled, the input media the workflow references is zipped beside the card too.

The library

The panel is a grid of cards over a folders + tags column, with an editable detail pane below.

Cards show the result poster (16:9), a P/N badge for which polarities exist, an "A" badge when input assets are attached, an "!" badge when a custom thumbnail is hidden, a media-type icon (🖼️ / 📽️ / 🔊 / 📖), the title, tags, and quick Fill / Copy buttons.
Single-click selects (instant); double-click the thumbnail opens the full image/video/audio; double-click the title renames in place; middle-click spawns both prompt nodes; drag onto a folder to move; drag onto the canvas to rebuild the original workflow; right-click for the full menu.
The toolbar has Search, the ▤ Details toggle, a sort button (↓ Newest → ↑ Oldest → A→Z → Z→A), a tag filter, a size slider for the grid, a 📦 Pack button, and ⟳ Refresh to rescan the library.

Folders & tags

Folders are real directories on disk. Tags live in a small registry so you can create empty tags up front and assign later — drag tags onto cards (or cards onto a tag) to tag in bulk. Click a tag to filter; the tag pane has its own size and sort. G-Prompts remembers the folder you left off in and reopens there (Settings → Remember selected folder).

Detail pane — edit positive & negative

Toggle ▤ Details to open resizable Positive and Negative editors (on by default on fresh installs). Edit the text and Save; each block has its own Fill and Copy. Edits are written straight to the card's files — and a card you edited then drag back onto the canvas rebuilds the workflow with your edited prompt, not the original.

Fill / re-inject

Fill pushes a card's prompt into your workflow: with text nodes selected it fills them by polarity (negative-detected node gets the negative, everything else the positive); with nothing selected it spawns fresh prompt nodes. Great for reusing a favourite prompt across new graphs.

Manage cards

Right-click any card:

Rename, edit/copy/paste/clear tags, Overwrite from current workflow (re-capture into this card, keeping its title + tags), Cut / Copy / Paste / Duplicate, add / replace / remove / hide the thumbnail, pick the default thumbnail style, Pack, and Move to Trash.

🗑 Trash — nothing is ever destroyed

Deleting a card, thumbnail, or folder moves it to a Trash/ folder inside your library, keeping its original sub-folder structure. The pinned Trash row in the folder pane shows how much space it's using; open it and right-click any card to Restore it to where it came from. G-Prompts contains no file-deletion code at all — emptying the Trash is something you do yourself in your file manager, when you're sure.

Pack / export

Pack zips the selected card(s) — prompt, metadata, media, workflow, and (optionally) captured input assets — and copies the archive's path to your clipboard, with a toast to confirm. Perfect for sharing a prompt + its result + the exact graph that made it.

Settings

Prompt Storage location — change where your library lives.
Capture shortcut — rebind the capture hotkey (default Alt+Q).
Auto-close panel — collapse G-Prompts when you click back into the canvas.
Remember selected folder — reopen in the folder you last selected.
Capture input assets — on capture, zip the workflow's referenced input media beside the card (.assets.zip).
Include captured assets in packs — bundle the .assets.zip into Packs.
Pack notification duration — how long the "Packed — path copied" toast stays up (1–15 s).

🎬 Feature demos

Fifteen ~10-second clips, one per feature.

Auto-close panel

Click back into the canvas and the panel gets out of your way — one toggle in Settings.

https://github.com/user-attachments/assets/0c104062-03b1-4dc2-af35-f99e6d02e351

Set the storage location

Pick the folder your library lives in — every card is plain files on disk, ready to back up or sync.

https://github.com/user-attachments/assets/db0d2afc-2f65-4b5a-8195-00b271f81a20

Capturing prompt cards

One click grabs the exact positive/negative prompt behind your last result — with a thumbnail and a copy of the workflow.

https://github.com/user-attachments/assets/05a552ee-0c36-4413-9817-76ec934eb3aa

Capture any workflow

No sampler? No problem — LTX, TTS, and custom graphs are traced back from their output nodes to the prompts that made the result.

https://github.com/user-attachments/assets/09861c0e-bdbe-4d80-8d9b-ea8c7daa361b

Capture videos, with previews

Video results get a compact poster (via ffmpeg) and play right on the card — double-click to open the full clip.

https://github.com/user-attachments/assets/ac93890a-ea6f-457a-bca8-270d1b4141e8

Default thumbnails

Hide a card's custom thumbnail or pick which built-in placeholder (Image / Video / Audio / Other) it shows.

https://github.com/user-attachments/assets/cfde5004-2cc1-45e7-943c-87f305c18fab

Create folders

Folders are real directories on disk — make them in the panel, drag cards in to organise.

https://github.com/user-attachments/assets/930a6d03-45bf-46eb-bd76-9cdb834b03e9

Create tags

Create tags up front, drag them onto cards (or cards onto tags) to assign in bulk, click to filter.

https://github.com/user-attachments/assets/3d0cc802-c5ab-41c9-9328-8c86546bc472

Drag & drop into the canvas

Drop a card onto the canvas and the workflow that produced it is rebuilt — with your edits, if you made any.

https://github.com/user-attachments/assets/8aea8e43-5b1b-43af-9e49-801cce46054b

Fill selected nodes

Select your text node(s) and hit Fill — the positive and negative land in the right nodes automatically, by polarity.

https://github.com/user-attachments/assets/96ae042c-4cdf-4f0f-9ee1-107337827194

Fill with no selection

Nothing selected? Fill spawns fresh prompt nodes ready to wire into any graph.

https://github.com/user-attachments/assets/7275cb8b-fda3-4f17-ba6f-b17dbb52edcf

Duplicate, edit & overwrite cards

Duplicate a card, edit its positive/negative in the detail pane, or overwrite it from the current workflow while keeping its title and tags.

https://github.com/user-attachments/assets/ec78f88d-2cf8-44b1-ab98-4191e62a6d14

UI controls

Search, the ▤ Details toggle, sort cycling, tag filtering, the grid size slider, and ⟳ Refresh — a tour of the toolbar.

https://github.com/user-attachments/assets/4cdd93fb-5830-468b-afed-47289efb331a

Non-destructive delete

"Delete" just moves the card to your library's Trash folder — restore it any time; G-Prompts never destroys a file.

https://github.com/user-attachments/assets/ac1688f5-2c64-4335-87ad-9f86490594ff

Pack complete cards for sharing

Pack zips prompt + metadata + media + workflow (+ captured input assets) into one archive and copies its path to your clipboard.

https://github.com/user-attachments/assets/af4896f3-17c2-4749-9667-52438940be99

How it stores things

The filesystem is the database. Under your chosen root, each card is a small set of sidecar files sharing one name:

<root>/<Folder>/<name>.txt            the positive prompt
<root>/<Folder>/<name>.neg.txt        the negative prompt
<root>/<Folder>/<name>.json           metadata (title, tags, source, type)
<root>/<Folder>/<name>.<ext>          the result media (image / video / audio)
<root>/<Folder>/<name>.thumb.jpg      the 16:9 card poster
<root>/<Folder>/<name>.workflow.json  the workflow, for canvas rebuild
<root>/<Folder>/<name>.assets.zip     captured input assets (optional)
<root>/Trash/…                        trashed cards, original hierarchy preserved

Because it's plain files, you can back it up, sync it, or reorganise it in your file manager and G-Prompts will follow along.

Compatibility

ComfyUI with a recent frontend (subgraph-flattened capture is tested on frontend 1.45+).
Works the same whether ComfyUI runs locally or on a remote/headless server (all storage is server-side; the UI is a normal sidebar).