Hent-ai is a system designed to enhance AI agent interactions by automatically classifying the emotion of bot responses using a Large Language Model (LLM) and attaching a corresponding visual emotion image to Discord messages. The term "Hent" is a coined word meaning "intent." It supports integration with both OpenClaw and Hermes Agent platforms.

The core methodology of Hent-ai involves two primary components: emotion classification from text and the generation/management of emotion-specific visual assets.

Emotion Classification:
Hent-ai employs an LLM to perform text-based emotion classification. For every response generated by the AI agent, the LLM analyzes the text to determine the underlying emotion expressed. Six distinct emotions are supported:

• Happy: Signifying success, completion, or celebration.
• Neutral: For general, informational, or default responses.
• Loyalty: Indicating acknowledgment, greeting, or attentiveness.
• Sorry: Conveying apology or acknowledging mistakes.
• Confused: Expressing uncertainty or questions.
• Focused: Denoting working, investigating, or debugging.

The system then matches the classified emotion to a pre-generated image, which is automatically attached to the Discord message, eliminating the need for the agent to explicitly include image tags.

Emotion Image Generation:
The system requires six unique images, one for each supported emotion. The recommended approach for creating these images is "Character + Reference-Based Generation," a technique leveraging image-to-image (img2img) capabilities of generative AI models (e.g., DALL-E, Midjourney, Stable Diffusion).

1. Base Character Generation: A single base character image is first created using any image generation tool. This character serves as the visual identity of the AI agent.
2. Emotion Variant Generation: The base character image is then used as a reference input for the image generator. For each of the six emotions, a specific text prompt (e.g., "Same character as the reference image, expressing [emotion]. Simple background, consistent art style.") is provided along with the base image to generate variants depicting the character expressing the desired emotion. Examples of emotional cues for prompts include:
   • Happy: "smiling, thumbs up, celebrating"
   • Neutral: "calm, relaxed, default expression"
   • Loyalty: "saluting, nodding, attentive"
   • Sorry: "apologetic, bowing, sheepish"
   • Confused: "head tilt, question mark, puzzled"
   • Focused: "concentrating, working, determined"
3. File Naming and Placement: Generated images are renamed according to their emotion (e.g., happy.png, neutral.png) and placed in a designated assets/ directory.

To ensure visual consistency and effectiveness, the following guidelines are provided for image creation:

• Maintain a consistent art style across all images.
• Use simple backgrounds to ensure readability as small thumbnails.
• Ensure emotions are visually distinct to make the image swap meaningful.
• Prefer square aspect ratios (1:1) for optimal Discord rendering.
• Keep file sizes under 500KB for fast uploads.
• Use PNG format for transparency and clean edges.

Integration with Agent's SOUL.md (Persona File):
Hent-ai's effectiveness relies on the AI agent's textual output implicitly conveying emotion, rather than explicit instructions. The agent's persona file (typically SOUL.md or equivalent) needs to be configured to facilitate this:

• Removal of Media Tags: Any existing instructions for the agent to output MEDIA: tags for images must be removed, as Hent-ai automates image attachment.
• Defining Emotional Behaviors: Instead of direct image commands, the SOUL.md should define clear behavioral patterns linked to emotions. For example, "When a task is completed successfully, celebrate briefly" signals a "happy" response; "When you make a mistake, own it immediately" signals "sorry." This allows the LLM classifier to infer emotion from the agent's natural language.
• Promoting Personality Range: Agents should be designed with a varied personality to avoid monotone responses, which would consistently result in "neutral" classifications. A broader range of expressions (excitement, frustration, curiosity) allows for more distinct emotional classifications.
• Plugin Acknowledgment: A simple note can be added to the SOUL.md informing the agent about the emotion-image plugin and its automatic handling of images.

Hent-ai is licensed under the MIT License.