<h1 id="fluxpseudonegative:-a-comfyui-node-for-converting-negative-prompts-to-positive-attributes-u2qdskm06">FluxPseudoNegative: A ComfyUI Node for Converting Negative Prompts to Positive Attributes</h1>FluxPseudoNegative is an advanced custom node for ComfyUI that converts negative prompts into positive ones. It's designed to enhance prompt engineering for image generation models that don't natively support negative prompts or where using negative prompts significantly increases generation time. So instead of hacking CFG we simply invert your negative words and find their antonyms!There are some promising techniques to get negative prompts working in Flux that will probably render this idea completely useless... But since I had already done the work I decided to complete at least a minimally viable node and ship it.<h2 id="features-lfbzpb4ui">Features</h2><ul><li>Multiple antonym-finding strategies:<ul><li>Custom phrase handling</li><li>WordNet</li><li>NLTK</li><li>Hugging Face Transformers</li></ul></li><li>Comprehensive phrase handling for multi-word concepts</li><li>Sentiment analysis for strength adjustment</li><li>Concept expansion using word embeddings</li><li>Optional ConceptNet integration for expanded antonyms</li><li>Multiple processing complexity levels: basic, advanced, expert</li><li>Optional LLM integration for unresolved terms or full prompt conversion</li><li>User-customizable antonyms and system prompts</li></ul><h2 id="caveats-syo3ssyg8">Caveats</h2><ul><li>Its not meant to be perfect! Its an imperfect solution to the issues of using CFG making the generation time double more or less.</li><li>For now I would NOT use conceptnet expansion. Its not working as intended.</li><li>A word like Gross has multiple meanings and the correct one cannot be inferred. ('disgusting' is one meaning, and can also mean the 'total' in reference to taxes for instance). This can result in unexpected return words</li><li>If there is a concept like "Brown Horse" in the negative it may not work well on that. Again, this doesn't work for everything. </li></ul><h2 id="installation-bbphd1bww">Installation</h2><ol><li>Install via Git option using ComfyUI Manager. Or Clone this repository into your ComfyUI <code>custom_nodes</code> directory:</li></ol><pre><code>git clone https://github.com/yourusername/ComfyUI-FluxPseudoNegativePrompt.git</code></pre><ol><li>Install the required dependencies:</li></ol><pre><code>pip install nltk textblob requests</code></pre><ol><li>Download the required NLTK data: (it is also done automatically when used)</li></ol><pre><code>import nltk
nltk.download('averaged_perceptron_tagger')
nltk.download('punkt')
nltk.download('wordnet')</code></pre><h2 id="usage-0jm5zvvg4">Usage</h2><ol><li>In the ComfyUI interface, look for the "Flux Pseudo Negative" node under the "prompt_processing" category.</li><li>Connect the node to your workflow:<ul><li>Input your negative prompt</li><li>Input a positive prompt to augment</li><li>Adjust the strength parameter (0.0 to 1.0)</li><li>Select the processing complexity</li><li>(Optional) Provide custom antonyms in the format "word:antonym" (one per line)</li><li>(Optional) Enable ConceptNet integration</li><li>(Optional) Enable LLM integration (full or fallback)</li><li>(Optional) Provide a custom system prompt for LLM integration</li></ul></li><li>The node will output:<ul><li>A modified positive prompt incorporating the converted negative concepts</li><li>(If LLM integration is enabled) An LLM input string for further processing</li></ul></li></ol><h2 id="parameters-w2g5flyc7">Parameters</h2><ul><li><code>negative_prompt</code>: The negative prompt to convert</li><li><code>positive_prompt</code>: An optional positive prompt to augment</li><li><code>strength</code>: The strength of the antonym influence (0.0 to 1.0)</li><li><code>complexity</code>: The processing complexity level (basic, advanced, expert)</li><li><code>custom_antonyms</code>: Optional custom antonym mappings</li><li><code>use_conceptnet</code>: Enable ConceptNet integration for concept expansion</li><li><code>use_llm_full</code>: Enable full LLM-based prompt conversion</li><li><code>use_llm_fallback</code>: Enable LLM-based fallback for unresolved terms</li><li><code>custom_system_prompt</code>: Custom system prompt for LLM integration</li></ul><h2 id="file-structure-4if565jic">File Structure</h2><ul><li><code>__init__.py</code>: Initializes the node for ComfyUI</li><li><code>FluxPseudoNegative.py</code>: Contains the main <code>FluxPseudoNegativeNode</code> class</li><li><code>flux_utils.py</code>: Contains the <code>PhraseHandler</code> class and <code>strength_map</code>.</li></ul><h2 id="customization-rkv3p3w6p">Customization</h2>You can customize the phrase mappings and strength map by modifying the <code>flux_utils.py</code> file. You can also load custom antonym dictionaries, in the text box or from a text file one per line. You can also specify a custom LLM system prompt to integrate your negative prompt into for conversion in a 3rd party LLM node (the node comes packaged with 3 already tested/validated).<h2 id="note-zqyuf1sbk">Note</h2>This node requires significant computational resources, especially when using advanced NLP features and models. Performance may vary based on your system capabilities and the complexity of the input prompts.

Files with SHA256: d13e5ba202a3263dae3b92fae3929ac559f57cecc7d9e785db052294d3865550