<h1 id="promptjson-node-for-comfyui-t8txow2zb">PromptJSON Node for ComfyUI</h1><a target="_blank" rel="ugc" href="https://github.com/NeuralSamurAI/ComfyUI-PromptJSON/blob/main/image.png?raw=true"><img src="https://github.com/NeuralSamurAI/ComfyUI-PromptJSON/raw/main/image.png?raw=true" alt="PromptJSON" /></a><h2 id="description-r9owtawjn">Description</h2>PromptJSON is a custom node for ComfyUI that structures natural language prompts and generates prompts for external LLM nodes in image generation workflows. It aids in creating consistent, schema-based image descriptions with support for various schema types.<h2 id="features-ghykwlcyx">Features</h2><ul><li>Structures natural language prompts according to multiple schema types</li><li>Generates prompts for external LLM nodes with multiple strategies</li><li>Supports negative prompts</li><li>Adjustable complexity for output detail</li><li>Custom schema support for fine-tuning prompt structure</li><li>Multiple prompt generation strategies (One Shot and Few Shot)</li><li>Support for various schema types: JSON, HTML, Key-Value, Attribute-Based, Visual Layer Breakdown, Compositional Grid, and Artistic Reference</li><li>Error handling for robust operation</li></ul><h2 id="installation-946vx126u">Installation</h2><ol><li>Navigate to your ComfyUI custom nodes directory.</li><li>Clone this repository or copy the <code>PromptJSON.py</code>, <code>prompt_templates.py</code>, and <code>__init__.py</code> files into a new directory named <code>ComfyUI-PromptJSON</code>.</li><li>Restart ComfyUI or refresh your workflow.</li></ol><h2 id="usage-y9w5yzr30">Usage</h2><ol><li>Find the "Prompt JSON" node in the "prompt_converters" category in ComfyUI.</li><li>Connect the following inputs:<ul><li><code>prompt</code>: Your main prompt text</li><li><code>negative_prompt</code>: Elements to avoid in the image</li><li><code>complexity</code>: A float value between 0.1 and 1.0 to adjust output detail</li><li><code>llm_prompt_type</code>: Choose between "One Shot" or "Few Shot"</li><li><code>schema_type</code>: Select from JSON, HTML, Key, Attribute-Based, Visual Layer Breakdown, Compositional Grid, or Artistic Reference</li><li><code>custom_schema</code> (optional): Custom schema string (format depends on chosen schema type)</li></ul></li><li>Use the outputs:<ul><li><code>system_prompt</code>: The system instructions for the LLM</li><li><code>user_prompt</code>: The main prompt for the LLM, including the structured input</li><li><code>negative_passthru</code>: The negative prompt passed through unchanged</li><li><code>schema</code>: The formatted schema used for structuring the prompt</li></ul></li></ol><h2 id="llm-prompt-types-1fvt6n7gb">LLM Prompt Types</h2><ul><li>One Shot: Provides a single example of a structured response based on the input prompt.</li><li>Few Shot: Provides multiple examples of structured responses at different complexity levels.</li></ul><h2 id="schema-types-4cmv4zy9o">Schema Types</h2><ol><li>JSON: Traditional JSON structure for detailed image descriptions.</li><li>HTML: HTML-like tag structure for organizing image elements.</li><li>Key: Key-value pair structure for straightforward descriptions.</li><li>Attribute-Based: List of attributes focusing on specific aspects of the image.</li><li>Visual Layer Breakdown: Structured description of background, midground, foreground, and focus elements.</li><li>Compositional Grid: 3x3 grid-based description of image elements.</li><li>Artistic Reference: Detailed breakdown of artistic elements like subject, composition, color palette, etc.</li></ol><h2 id="workflow-4h7rv1wuo">Workflow</h2><ol><li>The PromptJSON node takes the user's input and structures it according to the chosen schema type.</li><li>It generates a system prompt and user prompt based on the chosen LLM prompt type and schema.</li><li>The outputs can be sent to an external LLM node for further processing.</li><li>The external LLM node generates the final, detailed image description based on these prompts.</li></ol><h2 id="custom-schema-xsm0jldlm">Custom Schema</h2>The custom schema feature allows you to define the structure of the output. The format depends on the chosen schema type. For example, for JSON:<pre><code>{
 "scene": {
 "time_of_day": "string",
 "weather": "string",
 "location": "string"
 },
 "subjects": [
 {
 "type": "string",
 "description": "string",
 "position": "string"
 }
 ],
 "style": {
 "artistic_movement": "string",
 "color_palette": ["string"],
 "mood": "string"
 },
 "camera": {
 "angle": "string",
 "shot_type": "string"
 }
}</code></pre><h2 id="example-output-jgim8u12a">Example Output</h2>Input:<ul><li>prompt: "A serene lake at sunset with a lone fisherman in a small boat"</li><li>negative_prompt: "urban, city, crowds"</li><li>complexity: 0.7</li><li>llm_prompt_type: "Few Shot"</li><li>schema_type: "JSON"</li><li>custom_schema: (The JSON schema shown above)</li></ul>Output:<ol><li>System Prompt: (Contains instructions and examples for the LLM)</li><li>User Prompt: (Contains the structured input prompt and schema)</li><li>Negative Passthru: "urban, city, crowds"</li><li>Schema: (The formatted JSON schema)</li></ol><h2 id="notes-6rclrpjc1">Notes</h2><ul><li>The system prompt and user prompt are designed to guide an external LLM in generating a detailed, structured image description.</li><li>Experiment with different LLM prompt types and schema types to find the most effective combination for your specific use case.</li><li>Adjust the complexity parameter to control the level of detail in the generated prompts.</li><li>Custom schemas allow for fine-tuned control over the structure of the generated descriptions.</li></ul><h2 id="support-xsxywd1mt">Support</h2>For issues, feature requests, or contributions, please open an issue or pull request in the GitHub repository.

Files with SHA256: 2354fc85932a768ed723b6b513a396354dcfde012d6a158d0618be4227b3a10b