kuzco-sentiment

Analyze the sentiment or mood conveyed by an image with a local vision LLM. Wraps kuzco's tool_llm_image_sentiment() ellmer tool. Use when the task needs the emotional tone of an image, e.g. for conte

SKILL.md

GitHub

kuzco-sentiment

Analyze the sentiment or mood conveyed by an image using a local vision language model, exposed as an ellmer tool by the kuzco R package.

When to use

The user wants the emotional tone of an image: cheerful, somber, threatening, calm. Useful for content moderation triage, marketing asset selection, or photo curation. For factual descriptions of image contents, use kuzco-classification.

Setup

library(kuzco)
library(ellmer)

chat <- chat_ollama(model = "qwen2.5vl")
chat$register_tool(tool_llm_image_sentiment())

Tool arguments (LLM-controlled)

  • image (required): local file path to a jpeg, jpg, or png image.
  • additional_prompt (optional): extra instructions, e.g. "consider cultural context".
  • language (optional): output language, e.g. "English".

Result

Compact JSON text with image_sentiment, image_score, sentiment_description, and image_keywords. Lead with the sentiment label and score, then the description. Present sentiment as the model's reading, not as ground truth.

Example

User: "What mood does img/sunset.jpg convey?"

  1. Call kuzco_image_sentiment with image = "img/sunset.jpg".
  2. Report the sentiment, score, and description.