ComfyUI-Zero2JSON

ZeroPrompt Integration for FLUX2-JSON - Position-as-seed procedural text generation for structured prompt building.

Overview

Zero2JSON provides shadow nodes that augment the FLUX2-JSON system with deterministic, profile-based procedural text generation. Each FLUX2-JSON input field receives a parallel ZeroPrompt node designed to generate contextually appropriate text.

Core Philosophy: Not replace, but augment and extend - providing unparalleled precision in multi-dimensional orchestration of procedural prompt inputs with guardrails in the form of user curation.

Installation

1. Clone or copy this folder to your ComfyUI custom_nodes directory
2. Install requirements:

   pip install -r requirements.txt
   

3. Restart ComfyUI

Node Categories

Subject Nodes (Zero2JSON/Subject)

• 🎯 Z2J Subject Description - Generate detailed subject descriptions
• 📍 Z2J Subject Position - Generate spatial positioning
• 🏃 Z2J Subject Action - Generate dynamic actions
• 🧘 Z2J Subject Pose - Generate static poses

Scene Nodes (Zero2JSON/Scene)

• 🌍 Z2J Scene - Generate scene/environment descriptions
• 🖼️ Z2J Background - Generate background descriptions

Style Nodes (Zero2JSON/Style)

• 🎨 Z2J Style - Generate artistic style descriptions
• 😊 Z2J Mood - Generate mood/atmosphere descriptions

Lighting Nodes (Zero2JSON/Lighting)

• 💡 Z2J Lighting - Generate lighting setup descriptions

Camera Nodes (Zero2JSON/Camera)

• 📐 Z2J Camera Angle - Generate camera angle descriptions
• 📏 Z2J Camera Distance - Generate shot framing descriptions
• 🔍 Z2J Camera DoF - Generate depth of field descriptions
• 🎯 Z2J Camera Focus - Generate focus target descriptions

Composition Nodes (Zero2JSON/Composition)

• 📐 Z2J Composition - Generate composition rule descriptions

Utility Nodes (Zero2JSON/Utility)

• ℹ️ Z2J Profile Info - Display profile statistics
• 📦 Z2J Batch Generator - Generate multiple prompts
• 🔀 Z2J Seed Mixer - Combine multiple seeds

Common Inputs

All generation nodes share these inputs:

Input	Type	Description
seed	INT	World seed for deterministic generation (0 - 4294967295)
prompt_index	INT	Position in infinite prompt space (0 - 4294967295)
profile	DROPDOWN	Select vocabulary profile
prefix	STRING	Text to prepend to generated prompt
suffix	STRING	Text to append to generated prompt

Profile System

Profiles are JSON files in the profiles/ directory that define vocabulary and templates:

{
  "name": "Profile Name",
  "description": "What this profile generates",
  "version": "1.0.0",

  "templates": [
    "Template with {pool_a} and {pool_b}",
    "Another template: {pool_a}, {pool_b}"
  ],

  "pools": {
    "pool_a": ["option1", "option2", "option3"],
    "pool_b": ["choice1", "choice2", "choice3"]
  }
}

Default Profiles

Profile	Node	Combinations
subject_description_default.json	Z2J_SubjectDescription	~31M
subject_position_default.json	Z2J_SubjectPosition	~7K
subject_action_default.json	Z2J_SubjectAction	~11K
subject_pose_default.json	Z2J_SubjectPose	~27K
scene_default.json	Z2J_Scene	~32K
background_default.json	Z2J_Background	~5K
style_default.json	Z2J_Style	~19K
mood_default.json	Z2J_Mood	~41K
lighting_default.json	Z2J_Lighting	~91K
composition_default.json	Z2J_Composition	~5K
camera_angle_default.json	Z2J_CameraAngle	~8K
camera_distance_default.json	Z2J_CameraDistance	~972
camera_dof_default.json	Z2J_CameraDoF	~7K
camera_focus_default.json	Z2J_CameraFocus	~3K

Workflow Integration

Connect Zero2JSON nodes to FLUX2-JSON inputs:

[Z2J_SubjectDescription] → [FLUX2_SubjectCreator].description
[Z2J_SubjectPosition]    → [FLUX2_SubjectCreator].position
[Z2J_SubjectAction]      → [FLUX2_SubjectCreator].action
[Z2J_Scene]              → [FLUX2_PromptAssembler].scene
[Z2J_Style]              → [FLUX2_StyleSelector].custom_style
[Z2J_Lighting]           → [FLUX2_PromptAssembler].lighting
[Z2J_CameraAngle]        → [FLUX2_CameraRig].angle
...

Determinism Guarantee

For any fixed (seed, prompt_index, profile):

• Same hash values are always computed
• Same template is always selected
• Same components are always chosen
• Same output is always generated

This holds across different machines, Python versions (3.7+), operating systems, and execution times.

Creating Custom Profiles

1. Copy an existing profile from profiles/
2. Rename with appropriate prefix (e.g., subject_description_fantasy.json)
3. Modify templates and pools for your use case
4. The node will automatically discover the new profile

Profile Naming Convention

Profiles should follow the naming pattern: {category}_{name}.json

• subject_description_*.json - Subject description nodes
• subject_position_*.json - Subject position nodes
• scene_*.json - Scene nodes
• style_*.json - Style nodes
• lighting_*.json - Lighting nodes
• camera_angle_*.json - Camera angle nodes
• etc.

Technical Details

Hash Function

Uses xxhash32 for O(1) deterministic hashing:

prompt_hash(seed, prompt_idx, component_idx) → uint32

Combination Calculation

total = len(templates) × Π(len(pool_i))

For example, a profile with 5 templates and 4 pools of sizes [10, 20, 15, 8] would have:

5 × 10 × 20 × 15 × 8 = 120,000 unique combinations

License

MIT License

Credits

• Based on DJZ_ZeroPrompt methodology
• Designed for integration with ComfyUI-FLUX2-JSON