Militant Hitchhiker's Switchblade Pack (v1.3)

Militant Hitchhiker's Switchblade Pack is a set of multi-function custom nodes for ComfyUI, covering text processing, prompt enhancement, model analysis/quantisation, and advanced scheduling/sampling.

This release reflects Switchblade v1.3 (as initialised in the loader) and includes new nodes for GOD/ARC/ODE EMA, Flux quant, and a Groq-powered prompt enhancer.

Installation

1. Open a terminal/PowerShell.
2. Navigate to your ComfyUI custom_nodes directory.
3. Clone the repo:

   git clone https://github.com/MilitantHitchhiker/MilitantHitchhiker-SwitchbladePack.git

4. Restart ComfyUI.

Notes

• The Groq node will auto-install the groq Python package on first run if missing.
• Loaded modules are defined in __init__.py and a nodes.log file is written at startup for quick discovery.

Configuration (Groq)

If you plan to use the Groq API Prompt Enhancer, set your API key and optional default system prompt in modules/groq_config.json.

• api_key: your Groq API key
• system_prompt: optional default system prompt used unless overridden per-node

You can also provide an override_system_prompt input on the node to temporarily replace the default.

Nodes

Text Processing

Text Appender (TextAppender_v2)

Concatenate up to five text inputs; control how they’re joined and optionally append the result to a file in ComfyUI’s output directory.

• Inputs (optional): text1..text5, input_delimiter (e.g. \n), output_delimiter (e.g. \n), output_file ("none" to disable)
• Output: STRING (the concatenated text)
• Category: MilitantAI/Switchblade/Text Processing

Typical use

1. Provide any of text1..text5.
2. Use \n to represent newlines for delimiters.
3. Set output_file to a filename to append the result; otherwise it only returns the text.

Prompt Generator (Dictionary) (IntegratedRandomPromptGenerator)

Randomly selects one item from up to four dictionary files and joins them.

• Dictionaries folder: ComfyUI/input/Dictionaries (recursively scanned)
• Inputs: dict1_file..dict4_file (or "none"), enable_dict1..enable_dict4, dict1_delimiter..dict4_delimiter (default ;), output_delimiter (default , ), seed
• Output: STRING
• Category: MilitantAI/Switchblade/Text Processing

Tip: Use ; as the default item delimiter inside your .txt dictionaries.

Groq API Prompt Enhancer (GroqAPIPromptEnhancer)

Send text to Groq chat models to enhance prompts.

• Models: gemma2-9b-it, llama-3.1-8b-instant, llama-3.3-70b-versatile
• Inputs: model, text, optional override_system_prompt
• Output: STRING (enhanced text)
• Config file: modules/groq_config.json
• Category: MilitantAI/Switchblade/Text Processing

Model Insight and Utilities

Model Analyser (ModelAnalyserNode)

Analyses a model’s state dict and returns a JSON summary.

• Input: model
• Output: STRING (JSON: structure, size, blocks, dtypes, totals)
• Category: MilitantAI/Switchblade/Model Merging

Flux Quant Node (FluxQuantNode)

Converts the model’s tensors to a selected precision, analyses them, and saves the processed model to ComfyUI’s output directory.

• Inputs: model, precision (auto, float32, float16, bfloat16, float8_e5m2, float8_e4m3fn)
• Output: STRING (analysis JSON; also prints save status)
• Saved file name: <ckpt_or_unknown>_<precision>.safetensors
• Category: MilitantAI/Switchblade/Model Merging

Generation (Schedulers and Samplers)

ARC Scheduler (ARC Scheduler)

Deterministic, gently warped descendent schedule; returns SIGMAS suitable for advanced samplers.

• Inputs: model, steps, denoise
• Output: SIGMAS
• Category: MilitantAI/Switchblade/Generation

GOD Scheduler (Advanced)

Produces a baseline descending SIGMAS schedule (high → low) with denoise control.

• Inputs: model, steps, denoise
• Output: SIGMAS
• Category: MilitantAI/Switchblade/Generation

GOD Sampler (Advanced)

Returns a SAMPLER wrapping a GOD-Flow sampler with synergy-adapted step behaviour.

• Output: SAMPLER
• Category: MilitantAI/Switchblade/Generation

GOD Sampler (Advanced) Ext.

Like the GOD Sampler (Advanced) but with a god_strength control.

• Inputs: god_strength
• Output: SAMPLER
• Category: MilitantAI/Switchblade/Generation

ODE EMA Sampler

Detail-aware ODE sampler with EMA-stabilised cone shaping and late-step detail emphasis.

• Inputs: stats_downsample, detail_gain_max, detail_power, detail_schedule_power
• Output: SAMPLER
• Category: MilitantAI/Switchblade/Generation

Typical Wiring Examples

• Use GOD Scheduler (Advanced) or ARC Scheduler → connect to KSampler (Advanced) sigmas input.
• Use GOD Sampler (Advanced) or GOD Sampler (Advanced) Ext. → connect to SamplerCustom sampler input.
• Use ODE EMA Sampler → connect to SamplerCustom sampler input.
• Use Flux Quant Node and Model Analyser → connect their model input to your loaded model.
• Use Text Appender and Prompt Generator (Dictionary) for pre/post text tasks in your workflow.

Contributing

Contributions are welcome! Please open an issue or PR on the GitHub repository.

Changelog

v1.3

• Added: GOD Scheduler (Advanced), GOD Sampler (Advanced), GOD Sampler (Advanced) Ext., ODE EMA Sampler.
• Added: ARC Scheduler for deterministic ARC-warped schedules.
• Added: Flux Quant Node for precision conversion + analysis + save.
• Added: Groq API Prompt Enhancer with modules/groq_config.json config.
• Updated: Text Appender now uses input_delimiter and output_delimiter and writes to ComfyUI output when output_file is set.
• Updated: Prompt Generator (Dictionary) defaults to ; as item delimiter and scans ComfyUI/input/Dictionaries recursively.
• Renamed: arc_lr_scheduler.py → ARC_scheduler.py.
• Note: model_save.py (Save Flux Model v2) exists but is disabled in __init__.py in this release.

License

MIT