src.prompts.prompt_utils module

Functions for creating and formatting prompts for image and language models.

This module provides utilities to generate prompts for text-to-image models and language models by combining concepts with visual descriptions and formatting generation history.

src.prompts.prompt_utils.concept_image_prompt(prompt_text: str, concept: str) → str

Create an image generation prompt by combining text with visual descriptions.

Combines a base prompt text with a concept and adds random visual descriptions (angles and lighting) to guide the image generation model.

Parameters:

• prompt_text – Base prompt text for the image.

• concept – Concept name to include in the prompt.

Returns:

Complete prompt string for text-to-image generation.

src.prompts.prompt_utils.generate_prompt(concept_history: Mapping[str, float], generation_history: Sequence[str], prompt_path: str, top_k: int | None = None) → str

Generate a prompt for the language model by filling a template.

Creates a formatted prompt by filling a template with the concept history (sorted by score) and generation history. Optionally limits to top-k concepts.

Parameters:

• concept_history – Dictionary mapping concept names to their scores.

• generation_history – Sequence of generated concept names.

• prompt_path – Path to the prompt template file.

• top_k – Optional number of top-scoring concepts to include. If None, includes all concepts.

Returns:

Formatted prompt string with concept and generation history inserted.