ComfyUI Node: MiniMax H3 Structured Prompt Enhancer (Plan v2)

MiniMax H3 Structured Prompt Enhancer (Plan v2):

The MiniMaxH3PlanV2PromptEnhancer is a sophisticated node designed to enhance and refine structured prompts within the MiniMax H3 framework. Its primary purpose is to augment the quality and effectiveness of prompts by applying advanced enhancement techniques, which can be tailored to either preserve the original intent or encourage creative expansion. This node is particularly beneficial for AI artists seeking to generate more nuanced and contextually rich outputs, as it leverages both textual and visual analysis capabilities to inform the enhancement process. By integrating seamlessly with other components of the MiniMax H3 system, it ensures that the enhanced prompts are not only coherent but also aligned with the desired artistic or functional goals.

MiniMax H3 Structured Prompt Enhancer (Plan v2) Input Parameters:

clip

The clip parameter requires a generation-capable CLIP model, such as Qwen3-VL or Qwen3.5, which is essential for conditioning the prompt enhancement process. This parameter ensures that the node can effectively interpret and manipulate the input data to produce high-quality outputs. The CLIP model acts as a foundational element, providing the necessary context and understanding for the enhancement operations.

plan_context

The plan_context parameter is a compiled input from the MiniMax H3 Prompt Merge (Plan v2). It serves as the structural backbone for the enhancement process, providing the necessary context and framework within which the prompt enhancements are applied. This parameter is crucial for maintaining coherence and alignment with the original plan.

system_prompt

The system_prompt parameter allows for the customization of the node's behavior through editable prose. By default, it uses a built-in production-detail pass, but users can modify it to suit specific needs. This parameter influences the narrative style and detail level of the enhanced prompt, offering flexibility in how the final output is presented.

visual_analysis

The visual_analysis parameter determines whether the node will perform visual analysis on reference images and sampled video. It can be set to either "Text metadata only (no visual analysis)" or "Analyze reference images and sampled video." This choice impacts the depth and type of analysis conducted, with visual analysis providing richer context for the enhancement process.

enhancement_mode

The enhancement_mode parameter dictates the style of prompt enhancement, offering options such as "Intent-locked expansion" and "Creative expansion." This setting influences whether the enhancements strictly adhere to the original intent or explore more creative interpretations, allowing users to tailor the output to their specific artistic goals.

analysis_long_edge

The analysis_long_edge parameter specifies the maximum dimension for visual analysis, affecting the resolution and detail level of the analysis conducted on visual inputs. This parameter is important for optimizing the balance between analysis depth and computational efficiency.

video_analysis_fps

The video_analysis_fps parameter sets the frames per second rate for video analysis, determining how frequently frames are sampled for analysis. This setting impacts the granularity of the visual analysis and can be adjusted to suit the complexity of the video content.

max_analysis_frames

The max_analysis_frames parameter limits the number of frames analyzed from a video, with a default value of 8. This constraint helps manage computational resources while ensuring sufficient data is available for effective analysis.

max_new_tokens

The max_new_tokens parameter defines the maximum number of new tokens that can be generated during the enhancement process, with a default value of 1200. This setting controls the length and complexity of the enhanced prompt, balancing detail with conciseness.

sampling

The sampling parameter influences the randomness and diversity of the generated output, affecting how the node explores different enhancement possibilities. This parameter is key for achieving the desired level of variability in the enhanced prompt.

temperature

The temperature parameter adjusts the randomness of the output generation, with higher values leading to more diverse outputs. This setting is crucial for controlling the creativity and unpredictability of the enhanced prompt.

top_k

The top_k parameter limits the number of top predictions considered during generation, affecting the diversity and quality of the output. This setting helps refine the balance between exploration and precision in the enhancement process.

top_p

The top_p parameter, also known as nucleus sampling, determines the cumulative probability threshold for selecting predictions, influencing the diversity of the output. This parameter is important for managing the trade-off between creativity and coherence.

min_p

The min_p parameter sets a minimum probability threshold for predictions, ensuring that only sufficiently likely options are considered. This setting helps maintain the quality and relevance of the enhanced prompt.

repetition_penalty

The repetition_penalty parameter discourages the model from repeating the same tokens, promoting more varied and interesting outputs. This setting is essential for enhancing the originality and engagement of the prompt.

presence_penalty

The presence_penalty parameter penalizes the model for using tokens that have already appeared, encouraging the exploration of new ideas. This setting is crucial for fostering creativity and diversity in the enhanced prompt.

seed

The seed parameter sets the random seed for the generation process, ensuring reproducibility of results. This parameter is important for achieving consistent outputs across different runs.

thinking

The thinking parameter is a boolean that, when enabled, allows the node to engage in more complex reasoning during the enhancement process. This setting can enhance the depth and sophistication of the output.

clip_tail

The clip_tail parameter, when specified, connects a Generation Tail Loader to the CLIP model, enhancing its conditioning capabilities. This parameter is optional but can significantly impact the quality of the enhanced prompt.

offload_after_generation

The offload_after_generation parameter, with a default value of False, determines whether resources are offloaded after generation. This setting helps manage computational resources and can improve performance in resource-constrained environments.

MiniMax H3 Structured Prompt Enhancer (Plan v2) Output Parameters:

enhanced_prompt

The enhanced_prompt output provides the refined and augmented version of the original prompt, incorporating the enhancements applied by the node. This output is crucial for achieving the desired artistic or functional goals.

editable_prose

The editable_prose output offers a version of the enhanced prompt that can be further edited, providing flexibility for users to make additional adjustments. This output is important for tailoring the final result to specific needs.

enhanced_plan_context

The enhanced_plan_context output delivers the updated plan context, reflecting the changes and enhancements applied during the process. This output ensures coherence and alignment with the original plan.

base_system_prompt

The base_system_prompt output presents the initial system prompt used in the enhancement process, providing context for the changes made. This output is useful for understanding the starting point of the enhancements.

effective_system_prompt

The effective_system_prompt output shows the final system prompt after enhancements, highlighting the modifications and improvements made. This output is key for evaluating the impact of the enhancement process.

llm_prompt

The llm_prompt output provides the prompt as interpreted by the language model, offering insights into how the enhancements were applied. This output is valuable for understanding the model's perspective on the prompt.

enhancer_report

The enhancer_report output delivers a detailed report on the enhancement process, including insights and metrics that can inform future enhancements. This output is essential for evaluating the effectiveness and quality of the enhancements.

MiniMax H3 Structured Prompt Enhancer (Plan v2) Usage Tips:

• To achieve a balance between creativity and coherence, experiment with different enhancement_mode settings, such as switching between "Intent-locked expansion" and "Creative expansion" based on your artistic goals.
• Utilize the visual_analysis parameter to incorporate visual context into your enhancements, especially when working with multimedia content, to enrich the prompt with additional layers of meaning.
• Adjust the temperature and top_p parameters to fine-tune the randomness and diversity of the output, allowing for more varied and interesting enhancements.
• Use the offload_after_generation parameter to manage computational resources effectively, especially in environments with limited processing power.

MiniMax H3 Structured Prompt Enhancer (Plan v2) Common Errors and Solutions:

"A generation-capable CLIP input is required."

• Explanation: This error occurs when the clip parameter is not provided or is incompatible with the node's requirements.
• Solution: Ensure that a valid and compatible CLIP model, such as Qwen3-VL or Qwen3.5, is connected to the node.

"Choose a supported visual_analysis mode."

• Explanation: This error indicates that an unsupported value was provided for the visual_analysis parameter.
• Solution: Verify that the visual_analysis parameter is set to either "Text metadata only (no visual analysis)" or "Analyze reference images and sampled video."

"Invalid enhancement mode selected."

• Explanation: This error suggests that an unsupported enhancement mode was chosen.
• Solution: Ensure that the enhancement_mode parameter is set to either "Intent-locked expansion" or "Creative expansion," as these are the supported options.