ComfyUI Prompt Manager

A small node pack for ComfyUI:

• Prompt Manager — template/history-driven prompt supplier
• Presentation Image Saver — full-screen image grid viewer + save

Both nodes ship from this one repo/folder so ComfyUI only has to load a single custom_nodes package.

Node: Prompt Manager (utils/prompt)

Outputs: prompt (STRING), enhancer_instruction (STRING)

Wire prompt into your CLIP Text Encode node to drive it from one place.

Template placeholders

The prompt may contain placeholders:

• {character}, {scene}, {environment}, {genre}, {style} — replaced with a pick from the corresponding list in templates.json (in this node's folder). Add your own aspects by adding keys to that file; any {key} matching a key is resolved. Template entries may themselves contain placeholders (resolved up to 5 levels). Click the node's Edit Templates button (also available via right-click) to open a spreadsheet-style table — one column per category, one row per item index: tick a checkbox to enable/disable an item, edit its text directly in its cell, drag the ⠿ handle to reorder within a column, or add/delete whole categories/items.
• {option a|option b|option c} — inline options, one is picked.
• {page} — replaced with the current page counter (see Pages below).
• Unknown placeholders are left untouched.

Example:

{genre}, {character} in {scene}, {environment}, {highly detailed|minimalist}

Shuffling

Picks are driven by the shuffle_seed widget, independent of any generation seed elsewhere in the workflow. For each placeholder, the picked item's index is shuffle_seed % item count (counting only enabled items) — so item order controls the sequence (drag to reorder in the editor):

• shuffle_seed → increment: steps through each aspect's list in order, wrapping around — a predictable, repeatable sequence across runs.
• shuffle_seed → randomize: jumps to an arbitrary index each run.
• shuffle_seed → fixed: the current picks are kept exactly.

Pages

Set pages to N to keep the same resolved template picks for N consecutive runs (e.g. for a multi-page/panel sequence), using {page} in the prompt to print the current counter (1..N). After N runs the counter resets and picks reshuffle from shuffle_seed. Enable debug_print to print the final resolved prompt to the console each run.

History

Every executed run is saved to history.json (last 200, deduped, newest first). The history dropdown on the node lists past prompts; selecting an entry restores the prompt, pages, shuffle seed and enhancer into the node.

Enhancers

The enhancer dropdown selects a saved system/enhancer instruction (e.g. a prompt-rewriting instruction for an LLM text encoder) whose text is emitted as the enhancer_instruction output — wire it together with prompt into a "Generate Text" node to drive prompt enhancement. Manage the saved list (add, edit, delete) via the node's Edit Enhancers button (also available via right-click); entries are stored in enhancers.json (seeded with a "Comic Page Art Director" example). Selecting None outputs an empty string.

Since Generate Text's single prompt input expects the full chat-formatted text, combine the two outputs into the model's chat template yourself, e.g. for Qwen-style models via a StringConcatenate/StringFormat node chain into:

<|im_start|>system
{enhancer_instruction}<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant

API

• GET /prompt_manager/history — list entries
• POST /prompt_manager/history/delete — body {"timestamp": <ts>} deletes one entry; {} clears all
• GET / POST /prompt_manager/templates — read/replace the aspect lists ({aspect: [{"text": string, "enabled": bool}, ...]})
• GET / POST /prompt_manager/enhancers — read/replace the saved enhancer instructions ({name: instruction_text})

Node: Presentation Image Saver (image)

Input: images (IMAGE batch), layout (1x1/2x1/2x2/3x3), save_mode (auto/manual), prompt_text (STRING, optional)

Adds a Present (Full Screen) button to the node that opens a full-screen grid showing the images as they arrive. In manual save mode, each image gets a Save button to copy it from the temp directory into the output directory; in auto mode, images are saved to the output directory immediately and marked as already saved.

If prompt_text is non-empty, it's saved as a .txt file next to each image using the same filename (e.g. ComfyUI_00001_.png + ComfyUI_00001_.txt) — written at save time, so in manual mode this happens when you click Save, not when the image first appears. Wire Prompt Manager's prompt output (or any STRING) into it to keep the actual prompt alongside each saved image.

API

• POST /presentation_node/save — body {"filename", "subfolder", "type", "prompt_text"}, copies a temp image into the output directory (and its .txt companion, if prompt_text is set)