AF - Edit Generated Prompt - Receive LLM-generated prompts and optionally edit them manually
AF - Save Prompt History - Archive prompts to JSON files with timestamps
AF - Load Prompt History - Browse and load previous prompts by index
AF - Show Text - Simple node that displays the connected text
AF - Model Switch - A dynamic model/clip/vae switching node

🚀 Installation

Via ComfyUI Manager (Recommended)

Open ComfyUI Manager
Search for "AF - Pack Prompt Nodes"
Install

Manual Installation

cd ComfyUI/custom_nodes/
git clone https://github.com/alFrame/ComfyUI-AF-Pack-Prompt-Nodes.git

AF - Edit Generated Prompt

Purpose: Pipe LLM output through your workflow with optional manual editing.

How It Works

Upper field (read-only, grayed): Displays incoming prompt from LLM
Lower field (editable): Manual input or edited version
Output priority: Lower field → Upper field → Empty string

Key Feature: Click "Copy Generated Prompt for Editing" to move prompt from upper to lower field for modifications.

Typical Usage

[Ollama/LLM Node] → [AF - Edit Generated Prompt] → [CLIP Text Encode]

Download example workflow (right click -> save as...)

Short Video

https://github.com/user-attachments/assets/f16e0324-ddad-41fd-92a9-860cac2d02e3

Watch the Video on Youtube in higher Resolution

Video of "AF - Edit Generated Prompt" in a workflow

AF - Save Prompt History

Purpose: Save up to 4 prompts per entry with automatic timestamps.

Inputs

directory - Folder name (default: "Prompt-History")
filename - JSON file name (default: "default")
project - Project identifier (stored as metadata in JSON for reference)
global_positive/negative - Main prompts (optional)
local_positive/negative - Secondary prompts (optional)

Download example workflow (right click -> save as...)

Where Files Are Saved

Primary: ComfyUI/output/{directory}/ (user working files)
Fallback: {node_pack_folder}/{directory}/ (bundled defaults)

Output Format

[
  {
    "timestamp": "2025-10-29 14:32:15",
    "project": "MyProject",
    "global_positive": "your prompt",
    "global_negative": "negative prompt",
    "local_positive": "local detail",
    "local_negative": "local negative"
  }
]

Note: All prompts are automatically cleaned (trailing whitespace removed, internal formatting preserved). The project field is stored for your reference but not used for filtering.

🔗 Common Workflow Patterns

AI-Assisted with History Backup with the "AF - Save Prompt History" Node

[LLM] → [Edit Prompt] ┬ → [CLIP Encode] → [KSampler]
                      └ → [Save History]

AF - Load Prompt History

Purpose: Load saved prompts using index-based time-travel through your history.

Quick Reference

┌─────────────────────────────────┐
│ AF - Load Prompt History        │
├─────────────────────────────────┤
│ directory: [AF-PromptHistory  ] │
│ filename:  [MyProject        ▼] │
│ timestamp_index: [0           ] │
├─────────────────────────────────┤
│ Outputs:                        │
│ • global_positive               │
│ • global_negative               │
│ • local_positive                │
│ • local_negative                │
│ • info (connect to Show Text!)  │
└─────────────────────────────────┘

How It Works

Filename dropdown - Lists all .json files in the directory (without .json extension)
Index selection - Choose which timestamp to load (0 = newest)
Info output - Shows selected prompt content and metadata

Download example workflow (right click -> save as...)

Index System

0 = Most recent prompt
1 = Second most recent
2 = Third most recent
etc.

Directory Priority

Files in output/{directory}/ override files in {node_pack}/{directory}/ with the same name.

Critical: Connect the Info Output!

Always connect info → Show Text node to see:

Currently selected filename and timestamp
Full content of all 4 prompts being loaded
Total available timestamps count
Error messages if index is out of range

Example Info Display

══════════════════════════════════════════════════
  AF - LOAD PROMPT HISTORY
══════════════════════════════════════════════════
📄 Filename: MyProject
📊 Available Timestamps: 205

🔢 SELECTED: [126] | 2025-10-29 | 14:32:15

Global Positive:
A serene mountain landscape at sunset, golden hour lighting...

Global Negative:
ugly, deformed, low quality, blurry

Local Positive:
detailed foreground rocks, wildflowers

Local Negative:
(empty)

Console Output:

AF Load - ✓ Successfully loaded: ID: 126 | 2025-10-29 14:32:15

🔗 Common Workflow Patterns (suggestions)

"AF - Load Prompt History" Node, Compare Previous Prompts

[Load History: index=0] → [Edit] → [CLIP] ┐
[Load History: index=5] → [Edit] → [CLIP] ├→ [KSampler]
[Load History: index=10]→ [Edit] → [CLIP] ┘

Pattern 3: Multi-File Workflow

[Load: FileA.json] → [Edit Global+] → ┐
[Load: FileB.json] → [Edit Local+]  → ├→ [CLIP Encode]
[Load: FileC.json] → [Edit Global-] → ┘

💡 Pro Tips

Index 0 is your friend - Always points to newest prompt
Info output shows everything - See exactly what you loaded
Console logging - Check ID and timestamp confirmation
Multiple load nodes - Each can access different files/indices
Workflows persist - Your selections save with the workflow
Project field - Use it as a mental note when saving, great for manual inspection

AF - Show Text

A very simple node that shows the text connected to the input of the Node

🔧 Troubleshooting

Files not appearing in dropdown

Check ComfyUI console for cache messages
Verify .json files exist in output/{directory}/
Restart ComfyUI to rebuild cache
Ensure files end with .json extension

Index out of range

Connect info output to Show Text to see available count
Example: 205 timestamps = valid indices are 0-204
The info display shows total count at the top

Prompts not loading

Ensure filename is selected (not "none")
Check timestamp_index is within range (see info output)
Verify JSON file format matches example above
Look for error messages in console

Directory not found

Use relative path from ComfyUI root e.g.: AF-PromptHistory
Or absolute path: /full/path/to/directory
Check folder exists and has read/write permissions

AF - Model Switch

A dynamic model switching node that allows you to toggle between multiple model/CLIP/VAE combinations without rewiring your workflow.

<img width="1067" height="945" alt="image" src="https://github.com/user-attachments/assets/dab5fb34-0161-40ab-97c0-f7280a6ccc74" /> On the left the model switch node with it's 2 default inputs and set to auto (Any Switch node style). On the right with all 10 inputs enabled and rail selection set to use rail 3.

Features

Dynamic Rails: Add 2-10 model rails on demand
Smart AUTO Mode: Automatically selects the first connected rail
1-Based Indexing: Intuitive numbering (Rail 1, Rail 2, Rail 3...)
Permanent Base Rails: Rails 1-2 never lose connections when adjusting rail count
Real-Time Status: Color-coded display shows active rail instantly
Flexible Sizing: Resizable node that preserves your layout preferences

How It Works

The Model Switch acts as a gateway for your model pipeline. Instead of maintaining multiple workflow branches or manually reconnecting loaders, you can:

Connect multiple checkpoint loaders (or AIO loaders) to different rails
Choose which rail to use via the selected_rail parameter
The node outputs the MODEL, CLIP, and VAE from your selected rail

Parameters

| Parameter | Type | Range | Description | |-----------|------|-------|-------------| | number_of_rails | INT | 2-10 | Number of visible model rails | | selected_rail | INT | 0-10 | Active rail: 0 = AUTO (first connected), 1-10 = manual selection | | status | STRING | - | Read-only display showing current active rail |

Status Display Colors

🟠 Orange: AUTO mode - automatically using first connected rail
🔵 Blue: Manual mode - specific rail selected
🔴 Red: Error - no models connected

Usage Example

Basic Setup:

Checkpoint Loader A → model_1, clip_1, vae_1
Checkpoint Loader B → model_2, clip_2, vae_2
                          ↓
                   AF Model Switch
                 (selected_rail: 0)
                          ↓
                  → model, clip, vae

A/B Testing Workflow:

Set number_of_rails: 2
Connect two different models to Rails 1 and 2
Set selected_rail: 1 to use first model, 2 for second model
Set selected_rail: 0 for AUTO mode (uses first connected)

Advanced: Quality vs Speed:

Set number_of_rails: 3
Rail 1: High-quality slow model
Rail 2: Balanced model
Rail 3: Fast draft model
Switch between them by changing selected_rail

Tips

Rails 1-2 are permanent: These connections are never lost when you adjust number_of_rails
Rails 3-10 are dynamic: These are added/removed as needed
Use with rgthree Group Bypasser: Bypass entire model loading groups to save memory - only the selected rail's loaders will execute
AUTO mode is smart: It will find the first rail with actual data, skipping any bypassed/muted nodes
Status updates instantly: No need to run the workflow to see which rail is selected

Integration with Other Nodes

Works great with:

rgthree Context Switch - for full pipeline switching
rgthree Group Bypasser - to disable unused model loaders
AF - Save Prompt History - save which model was used with each prompt
Any checkpoint loader or AIO loader node

Technical Notes

Input names are 1-based: model_1, clip_1, vae_1, etc.
Internally consistent: no confusing 0-based conversions
Minimum node width: 160px
Height auto-adjusts based on number of rails

📋 Requirements

ComfyUI (recent version recommended)
No external dependencies required
Works with any LLM/text generation nodes, or your manual prompting text nodes

📄 License

MIT License - See LICENSE file for details

🐛 Issues & Contributions

Found a bug? Have a feature request?
→ GitHub Issues

⚠️ Disclaimer

This ComfyUI custom node is developed through AI-assisted coding. While carefully tested, it is provided "as is" without warranty.

By using this node pack:

You install and run at your own risk
The creator is not liable for damages or data loss
Compatibility with your setup is not guaranteed
Test in a safe environment before production use

Report issues on GitHub - we appreciate your feedback!

📚 Additional Resources

Changelog - Version history and updates
GitHub Issues - Report bugs & request features
License - MIT License details

Made with ❤️ by Alex Furer & Qwen3, Claude AI, DeepSeek