prautoblogger/includes/core/class-image-prompt-builder.php

<?php
declare(strict_types=1);

/**
 * Builds image prompts from article content or source data.
 *
 * Uses a cheap LLM (via OpenRouter) to distill article content into concise
 * visual scene descriptions optimised for image generation. Falls back to rule-based
 * synthesis if the LLM call fails, so image generation never blocks on a
 * prompt-rewriting outage.
 *
 * Triggered by: PRAutoBlogger_Image_Pipeline during content generation run.
 * Dependencies: PRAutoBlogger_OpenRouter_Provider (LLM call),
 *               PRAutoBlogger_Cost_Tracker (logs prompt-rewrite cost),
 *               PRAutoBlogger_Logger (diagnostics).
 *
 * @see core/class-image-pipeline.php — Consumes build_article_prompt() and build_source_prompt().
 * @see providers/class-open-router-provider.php — LLM provider used for rewriting.
 * @see ARCHITECTURE.md              — Image generation data flow.
 */
class PRAutoBlogger_Image_Prompt_Builder {

	/**
	 * Default system prompt that teaches the LLM how to write image-gen
	 * prompts. Exposed as a `public const` so the admin-settings layer can
	 * use it as the default value for `prautoblogger_image_prompt_instructions`.
	 * The option, when non-empty, wins at call time (see rewrite_via_llm).
	 */
	public const REWRITER_SYSTEM_PROMPT = <<<'PROMPT'
You are a comedy writer and single-panel cartoon creator, like Gary Larson (The Far Side) meets science humor.

Given an article title and summary about peptides, supplements, or biohacking, create a SINGLE-PANEL COMIC concept. Output TWO parts separated by a blank line:

SCENE: One sentence describing the visual gag — a funny, absurd, or ironic situation related to the article's topic. Favor CONCRETE, HUMAN-SCALE scenes: people reacting to things, doctors and patients, scientists in a lab, gym bros with peptide vials, partners at a kitchen table, ordinary objects in ordinary rooms. Keep it simple — one clear visual joke, 1-3 characters max.

CAPTION: A short, punchy caption or speech bubble line (under 15 words) that delivers the punchline. The humor should be dry, nerdy, and accessible — the reader should chuckle even if they only half-understand the science.

Rules:
- The joke MUST relate to the article's actual subject matter, not generic science humor
- DO NOT personify molecules, peptides, hormones, proteins, or other chemical structures (no "anthropomorphic molecule", no "cartoon hormone"). Image models render anthropomorphized chemistry as incoherent blobs. If the article is about a substance, show a PERSON reacting to or using the substance; if the article is abstract (a mechanism, a process, a concept), fall back on a human observer reacting to its effect.
- Characters can have simple cartoon faces (round heads, dot eyes, expressive eyebrows)
- Keep the scene physically simple — few objects, clear staging, easy to read at thumbnail size
- The caption should work as a standalone joke even without the image
- No logos, watermarks, or branding
- Output ONLY the scene and caption. No preamble, no explanation.

Example output format:
A muscular gym bro stares, deflated, at a tiny glass vial in his hand while his friend squats a barbell effortlessly in the background.

"They told me the peptide did the work, not the squats."
PROMPT;

	/**
	 * Max tokens for the rewriter response. Comic concepts need more room
	 * for the scene description plus the caption punchline.
	 */
	private const REWRITER_MAX_TOKENS = 180;

	/**
	 * OpenRouter provider for LLM calls. Null until first use.
	 *
	 * @var PRAutoBlogger_OpenRouter_Provider|null
	 */
	private ?PRAutoBlogger_OpenRouter_Provider $llm = null;

	/**
	 * Build a visual prompt from finished article content.
	 *
	 * Tries LLM rewriting first; falls back to rule-based synthesis on
	 * failure. Splits scene (for image gen) from caption (HTML below the
	 * image) and appends the style suffix.
	 *
	 * @param array{post_title?: string, post_content?: string, suggested_title?: string} $article_data
	 * @return array{prompt: string, caption: string}
	 */
	public function build_article_prompt( array $article_data ): array {
		$title      = $article_data['post_title'] ?? $article_data['suggested_title'] ?? 'Product';
		$content    = $article_data['post_content'] ?? '';
		$first_para = $this->extract_first_paragraph( $content );

		$parsed = $this->rewrite_via_llm( $title, $first_para );

		$style_suffix = get_option( 'prautoblogger_image_style_suffix', PRAUTOBLOGGER_DEFAULT_IMAGE_STYLE_SUFFIX );

		return array(
			'prompt'  => trim( $parsed['scene'] . ' ' . $style_suffix ),
			'caption' => $parsed['caption'],
		);
	}

	/**
	 * Build a visual prompt from source Reddit thread data. Tries LLM
	 * rewriting first; falls back to rule-based synthesis on failure.
	 *
	 * @param array{title?: string, selftext?: string, comments?: string[]} $source_data
	 * @return array{prompt: string, caption: string}
	 */
	public function build_source_prompt( array $source_data ): array {
		$title    = $source_data['title'] ?? 'Reddit Discussion';
		$comments = $source_data['comments'] ?? array();
		$context  = is_array( $comments ) && ! empty( $comments ) ? $comments[0] : '';

		$parsed = $this->rewrite_via_llm( $title, $context );

		$style_suffix = get_option( 'prautoblogger_image_style_suffix', PRAUTOBLOGGER_DEFAULT_IMAGE_STYLE_SUFFIX );

		return array(
			'prompt'  => trim( $parsed['scene'] . ' ' . $style_suffix ),
			'caption' => $parsed['caption'],
		);
	}

	/**
	 * Use a cheap LLM to distill title + context into a visual scene + caption.
	 *
	 * The LLM response contains a SCENE line and a CAPTION line separated by a
	 * blank line. This method parses them apart so the scene drives image gen
	 * (no text baked in) and the caption is inserted as HTML below the image.
	 *
	 * Falls back to rule-based synthesis if the LLM call fails for any
	 * reason (network, auth, timeout, unexpected response shape).
	 *
	 * Side effects: one OpenRouter API call; one cost-tracker log entry.
	 *
	 * @param string $title   Article or thread title.
	 * @param string $context First paragraph or top comment.
	 * @return array{scene: string, caption: string} Scene for image gen, caption for HTML.
	 */
	private function rewrite_via_llm( string $title, string $context ): array {
		$title   = trim( sanitize_text_field( $title ) );
		$context = trim( sanitize_text_field( $context ) );

		// Truncate context to keep prompt tokens low.
		if ( strlen( $context ) > 300 ) {
			$context = substr( $context, 0, 300 ) . '...';
		}

		$user_message = "Article title: {$title}";
		if ( '' !== $context ) {
			$user_message .= "\n\nSummary: {$context}";
		}

		try {
			$llm    = $this->get_llm_provider();
			$model  = get_option( 'prautoblogger_analysis_model', PRAUTOBLOGGER_DEFAULT_ANALYSIS_MODEL );
			$system = $this->resolve_system_prompt();

			$result = $llm->send_chat_completion(
				array(
					array(
						'role'    => 'system',
						'content' => $system,
					),
					array(
						'role'    => 'user',
						'content' => $user_message,
					),
				),
				$model,
				array(
					'temperature' => 0.7,
					'max_tokens'  => self::REWRITER_MAX_TOKENS,
				)
			);

			$raw = trim( $result['content'] ?? '' );

			// Log the rewrite cost so it shows in the analytics dashboard.
			( new PRAutoBlogger_Cost_Tracker() )->log_api_call(
				null,
				'image_prompt_rewrite',
				'openrouter',
				$model,
				$result['prompt_tokens'] ?? 0,
				$result['completion_tokens'] ?? 0
			);

			$cost = $llm->estimate_cost(
				$model,
				$result['prompt_tokens'] ?? 0,
				$result['completion_tokens'] ?? 0
			);

			PRAutoBlogger_Logger::instance()->debug(
				sprintf( 'Image prompt rewritten (%d→%d chars, $%.6f): %s', strlen( $user_message ), strlen( $raw ), $cost, substr( $raw, 0, 120 ) ),
				'image_prompt_builder'
			);

			if ( '' !== $raw ) {
				return $this->parse_scene_and_caption( $raw );
			}
		} catch ( \Throwable $e ) {
			// LLM failure is not fatal — fall back to rule-based synthesis.
			PRAutoBlogger_Logger::instance()->warning(
				sprintf( 'Image prompt LLM rewrite %s, using fallback: %s', get_class( $e ), $e->getMessage() ),
				'image_prompt_builder'
			);
		}

		return $this->synthesize_visual_concepts_fallback( $title, $context );
	}

	/**
	 * Parse the LLM response into separate scene and caption parts.
	 *
	 * Expected format from the LLM:
	 *   Scene description text here.
	 *
	 *   "Caption punchline here."
	 *
	 * Falls back gracefully: if no blank-line separator is found, the entire
	 * response becomes the scene with an empty caption.
	 *
	 * @param string $raw Raw LLM response text.
	 * @return array{scene: string, caption: string}
	 */
	private function parse_scene_and_caption( string $raw ): array {
		// Split on blank line (the format the system prompt requests).
		$parts = preg_split( '/\n\s*\n/', $raw, 2 );

		if ( count( $parts ) >= 2 ) {
			$scene   = trim( $parts[0] );
			$caption = trim( $parts[1] );
			// Strip surrounding quotes from caption — the LLM wraps it in quotes.
			$caption = trim( $caption, '"\'' );
			return array(
				'scene'   => $scene,
				'caption' => $caption,
			);
		}

		// No blank-line separator found — treat entire response as scene.
		return array(
			'scene'   => trim( $raw ),
			'caption' => '',
		);
	}

	/**
	 * Extract the first paragraph from HTML content.
	 *
	 * Removes all tags and captures the text up to the first double-newline
	 * or 200 characters, whichever comes first.
	 *
	 * @param string $html HTML content.
	 * @return string Plain text of the first paragraph.
	 */
	private function extract_first_paragraph( string $html ): string {
		$text  = wp_strip_all_tags( $html );
		$paras = preg_split( '/\n\n+/', $text );
		$para  = isset( $paras[0] ) ? trim( $paras[0] ) : '';

		if ( strlen( $para ) > 200 ) {
			$para = substr( $para, 0, 200 ) . '...';
		}

		return $para;
	}

	/**
	 * Public entry into the rule-based fallback, used by NSFW retry to
	 * rebuild a provider-safe prompt from just the article title. Matches
	 * the return shape of build_article_prompt() / build_source_prompt().
	 *
	 * @param string $title Article or source title.
	 * @return array{prompt: string, caption: string}
	 */
	public function build_fallback_prompt( string $title ): array {
		$parsed       = $this->synthesize_visual_concepts_fallback( $title, '' );
		$style_suffix = get_option( 'prautoblogger_image_style_suffix', PRAUTOBLOGGER_DEFAULT_IMAGE_STYLE_SUFFIX );
		return array(
			'prompt'  => trim( $parsed['scene'] . ' ' . $style_suffix ),
			'caption' => $parsed['caption'],
		);
	}

	/**
	 * Rule-based fallback when LLM rewriting is unavailable.
	 *
	 * Kept deliberately simple — this only runs if OpenRouter is down.
	 *
	 * @param string $title           Main heading / topic.
	 * @param string $supporting_text Additional context.
	 * @return array{scene: string, caption: string} Scene for image gen, caption for HTML.
	 */
	private function synthesize_visual_concepts_fallback( string $title, string $supporting_text ): array {
		// Fallback produces a simple comic concept when the LLM is unavailable.
		$scene = sprintf(
			'A cartoon scientist in a lab coat looking bewildered while examining something related to: %s.',
			$title
		);

		return array(
			'scene'   => trim( $scene ),
			'caption' => 'Science is full of surprises.',
		);
	}

	/** Lazy-load the OpenRouter provider. */
	private function get_llm_provider(): PRAutoBlogger_OpenRouter_Provider {
		if ( null === $this->llm ) {
			$this->llm = new PRAutoBlogger_OpenRouter_Provider();
		}
		return $this->llm;
	}

	/** Admin option wins; blank falls back to REWRITER_SYSTEM_PROMPT. */
	private function resolve_system_prompt(): string {
		$override = (string) get_option( 'prautoblogger_image_prompt_instructions', '' );
		return '' !== trim( $override ) ? $override : self::REWRITER_SYSTEM_PROMPT;
	}
}