ComfyUI Node: MiniMax H3 Prompt Enhancer (Legacy Qwen LLM)

MiniMax H3 Prompt Enhancer (Legacy Qwen LLM):

The MiniMaxH3PromptEnhancer is a sophisticated tool designed to refine and enhance prompts for the MiniMax H3 system, leveraging the capabilities of a generative Qwen CLIP model. This node serves as a crucial step in the creative process, transforming initial prompt drafts into polished, structured prompts that are ready for advanced AI processing. By integrating with the MiniMax H3 framework, it ensures that the prompts are not only enhanced in terms of language and structure but also aligned with the specific requirements of the H3 system. The node is particularly beneficial for users looking to generate high-quality audiovisual content, as it can analyze and enhance prompts based on a variety of media inputs, including images and videos. Its primary goal is to provide a seamless and efficient way to produce prompts that are both creative and technically sound, ensuring that the final output meets the desired artistic and technical standards.

MiniMax H3 Prompt Enhancer (Legacy Qwen LLM) Input Parameters:

clip

The clip parameter requires a connection to a complete instruction-tuned Qwen3-VL/Qwen3.5 model or MiniMax H3's normal 50-layer conditioning CLIP. This input is crucial as it provides the foundational model that the enhancer uses to process and refine the prompts. The parameter is designed to work with an LLM-capable ComfyUI CLIP, not an OpenAI CLIP vision model, ensuring compatibility with the MiniMax H3 system. The clip input is essential for the node's operation, as it dictates the model's behavior and influences the quality of the enhanced prompts.

manual_prompt

The manual_prompt parameter allows you to input a pre-LLM structured draft or a manually crafted H3 prompt. This parameter is vital for providing the initial content that the enhancer will refine. It is recommended to include detailed and creative descriptions in this input, as empty or default descriptions may not provide the Qwen model with sufficient material to enhance effectively. The manual_prompt serves as the starting point for the enhancement process, and its quality directly impacts the final output.

min_p

The min_p parameter is a float value that determines the threshold for dropping tokens with low probability relative to the best token. It ranges from 0.0 to 1.0, with a default value of 0.05. This parameter is used only when sampling is set to "sample" and helps in filtering out less likely tokens, ensuring that the generated prompt is coherent and relevant. A conservative setting like 0.05 is recommended to maintain the quality of the output.

repetition_penalty

The repetition_penalty parameter is a float value that discourages the repetition of phrases in the generated prompt. It ranges from 0.0 to 5.0, with a default value of 1.05. This parameter is crucial for maintaining the structural integrity of the prompt, as excessive repetition can lead to redundancy and loss of meaning. However, it is important to keep this value close to 1.0, as the H3 system deliberately repeats certain fixed labels, and an excessive penalty could disrupt the required structure.

MiniMax H3 Prompt Enhancer (Legacy Qwen LLM) Output Parameters:

enhanced_prompt

The enhanced_prompt is the cleaned and refined version of the candidate text produced by the Qwen model. This output is crucial as it represents the final, polished prompt that can be used for further processing or connected to an official MiniMax H3 conditioning node. It is important to review the enhancer report for any structural warnings before using this output.

system_prompt

The system_prompt provides the exact resolved system instructions used for the enhancement process. This output is useful for inspecting and customizing the behavior of the enhancement process. By connecting it to a text viewer, you can review and edit the system instructions to better suit your needs.

llm_prompt

The llm_prompt is a textual representation of the Qwen chat template, including system, inventory, and user turns. This output is intended for text debugging purposes only and should not be used as an H3 video prompt. It provides insight into the internal workings of the enhancement process and can be useful for troubleshooting and refining the prompt.

enhancer_report

The enhancer_report provides a detailed account of the resolved H3 output family, generation status, and any H3 structure warnings. This output is essential for understanding the success of the enhancement process, any conditioning-only skips, historical system-prompt upgrades, or fallback scenarios. It serves as a comprehensive report on the enhancement process and helps in identifying areas for improvement.

MiniMax H3 Prompt Enhancer (Legacy Qwen LLM) Usage Tips:

• Ensure that the manual_prompt is detailed and creative to provide the Qwen model with sufficient material for enhancement.
• Review the enhancer_report for any structural warnings before using the enhanced_prompt in further processing.
• Customize the system_prompt to better align the enhancement process with your specific needs and artistic goals.

MiniMax H3 Prompt Enhancer (Legacy Qwen LLM) Common Errors and Solutions:

"conditioning-only" issue

• Explanation: This error occurs when the clip is configured for conditioning-only, which means it lacks the necessary components for full prompt generation.
• Solution: Ensure that the clip is connected to a complete instruction-tuned Qwen3-VL/Qwen3.5 model or MiniMax H3's normal 50-layer conditioning CLIP. Verify that the clip is not set to conditioning-only mode.

Empty or repetitive output

• Explanation: This issue arises when the generated prompt is either empty or contains excessive repetition, which can occur due to inappropriate settings of the repetition_penalty or insufficient input in the manual_prompt.
• Solution: Adjust the repetition_penalty to a value closer to 1.0 to prevent excessive repetition. Ensure that the manual_prompt is detailed and provides enough content for the model to enhance effectively.