| 1 |
<?php |
| 2 |
/** |
| 3 |
* Vision client — describes an image with a multimodal model. |
| 4 |
* |
| 5 |
* Kept separate from the text clients on purpose. Every provider expresses |
| 6 |
* image input differently (OpenAI nests typed content parts, Anthropic wants a |
| 7 |
* base64 `source` block, Gemini wants `inline_data`), and folding three |
| 8 |
* incompatible shapes into `generate_completion()` would make the text path |
| 9 |
* harder to read for no benefit. This class owns that divergence and returns |
| 10 |
* one thing: a sentence describing the picture. |
| 11 |
* |
| 12 |
* Images are sent as base64 rather than by URL because a WordPress upload is |
| 13 |
* frequently unreachable from the provider — local sites, staging behind auth, |
| 14 |
* intranets, or a CDN that blocks bots (see the ChatGPT/User-Agent block in the |
| 15 |
* MCP threads). A URL that works for us often 403s for them. |
| 16 |
* |
| 17 |
* @package ThinkRank\AI |
| 18 |
* @since 1.28.0 |
| 19 |
*/ |
| 20 |
|
| 21 |
declare(strict_types=1); |
| 22 |
|
| 23 |
namespace ThinkRank\AI; |
| 24 |
|
| 25 |
use ThinkRank\Core\Settings; |
| 26 |
|
| 27 |
if (!defined('ABSPATH')) { |
| 28 |
exit; |
| 29 |
} |
| 30 |
|
| 31 |
/** |
| 32 |
* Multimodal image description. |
| 33 |
*/ |
| 34 |
class Vision_Client { |
| 35 |
|
| 36 |
/** |
| 37 |
* Largest image we will upload, in bytes. |
| 38 |
* |
| 39 |
* Providers cap request size (Anthropic ~5MB per image) and a photo |
| 40 |
* straight off a phone routinely exceeds it. WordPress already generates |
| 41 |
* scaled sizes, so we prefer one of those and only fall back to the |
| 42 |
* original when nothing smaller exists. |
| 43 |
*/ |
| 44 |
private const MAX_IMAGE_BYTES = 3500000; |
| 45 |
|
| 46 |
/** |
| 47 |
* Preferred registered image sizes, smallest adequate first. A description |
| 48 |
* does not need a 4000px original — 'medium_large' is plenty and keeps the |
| 49 |
* request (and the bill) small. |
| 50 |
* |
| 51 |
* @var string[] |
| 52 |
*/ |
| 53 |
private const PREFERRED_SIZES = ['medium_large', 'large', 'medium']; |
| 54 |
|
| 55 |
/** |
| 56 |
* Settings accessor. |
| 57 |
* |
| 58 |
* @var Settings |
| 59 |
*/ |
| 60 |
private Settings $settings; |
| 61 |
|
| 62 |
/** |
| 63 |
* Constructor. |
| 64 |
* |
| 65 |
* @param Settings|null $settings Settings instance. |
| 66 |
*/ |
| 67 |
public function __construct(?Settings $settings = null) { |
| 68 |
$this->settings = $settings ?? Settings::instance(); |
| 69 |
} |
| 70 |
|
| 71 |
/** |
| 72 |
* Providers that can accept an image, mapped to the models we send. |
| 73 |
* |
| 74 |
* @return array<string, string> provider => default vision model. |
| 75 |
*/ |
| 76 |
public static function capable_providers(): array { |
| 77 |
return [ |
| 78 |
'openai' => 'gpt-5-mini', |
| 79 |
'claude' => 'claude-sonnet-5', |
| 80 |
'gemini' => 'gemini-2.5-flash', |
| 81 |
]; |
| 82 |
} |
| 83 |
|
| 84 |
/** |
| 85 |
* Whether the configured provider can describe images. |
| 86 |
* |
| 87 |
* @return bool |
| 88 |
*/ |
| 89 |
public function is_available(): bool { |
| 90 |
$provider = (string) $this->settings->get('ai_provider', 'openai'); |
| 91 |
|
| 92 |
return isset(self::capable_providers()[$provider]) |
| 93 |
&& '' !== $this->api_key_for($provider); |
| 94 |
} |
| 95 |
|
| 96 |
/** |
| 97 |
* The stored key for a provider. |
| 98 |
* |
| 99 |
* Keys are held per provider (`openai_api_key`, `claude_api_key`, |
| 100 |
* `gemini_api_key`) — the same names AI_Manager reads. There is no |
| 101 |
* provider-agnostic `ai_api_key` setting, so asking for one always |
| 102 |
* returned '' and made every vision call fall back to the filename |
| 103 |
* template without ever reaching the provider. |
| 104 |
* |
| 105 |
* @param string $provider Provider slug. |
| 106 |
* @return string |
| 107 |
*/ |
| 108 |
private function api_key_for(string $provider): string { |
| 109 |
return trim((string) $this->settings->get($provider . '_api_key', '')); |
| 110 |
} |
| 111 |
|
| 112 |
/** |
| 113 |
* Describe an attachment in one short sentence suitable for alt text. |
| 114 |
* |
| 115 |
* @param int $attachment_id Attachment to describe. |
| 116 |
* @param string $context Optional context (post title) to disambiguate. |
| 117 |
* @return string Alt text, or '' when it could not be produced. |
| 118 |
* @throws \Exception When the provider is unusable or the call fails. |
| 119 |
*/ |
| 120 |
public function describe_attachment(int $attachment_id, string $context = ''): string { |
| 121 |
$provider = (string) $this->settings->get('ai_provider', 'openai'); |
| 122 |
$models = self::capable_providers(); |
| 123 |
|
| 124 |
if (!isset($models[$provider])) { |
| 125 |
throw new \Exception(sprintf( |
| 126 |
/* translators: %s: AI provider name. */ |
| 127 |
esc_html__('The %s provider cannot describe images. Switch to OpenAI, Claude or Gemini in Settings → AI Provider.', 'thinkrank'), |
| 128 |
esc_html($provider) |
| 129 |
)); |
| 130 |
} |
| 131 |
|
| 132 |
$api_key = $this->api_key_for($provider); |
| 133 |
if ('' === $api_key) { |
| 134 |
throw new \Exception(esc_html__('No AI API key configured.', 'thinkrank')); |
| 135 |
} |
| 136 |
|
| 137 |
$image = $this->read_image($attachment_id); |
| 138 |
if (null === $image) { |
| 139 |
throw new \Exception(esc_html__('The image file could not be read, or is larger than the provider allows.', 'thinkrank')); |
| 140 |
} |
| 141 |
|
| 142 |
$model = (string) $this->settings->get('ai_vision_model', $models[$provider]); |
| 143 |
$prompt = $this->build_prompt($context); |
| 144 |
|
| 145 |
switch ($provider) { |
| 146 |
case 'openai': |
| 147 |
$text = $this->call_openai($api_key, $model, $prompt, $image); |
| 148 |
break; |
| 149 |
case 'claude': |
| 150 |
$text = $this->call_claude($api_key, $model, $prompt, $image); |
| 151 |
break; |
| 152 |
default: |
| 153 |
$text = $this->call_gemini($api_key, $model, $prompt, $image); |
| 154 |
break; |
| 155 |
} |
| 156 |
|
| 157 |
return $this->clean($text); |
| 158 |
} |
| 159 |
|
| 160 |
/** |
| 161 |
* The instruction. Alt text has rules — screen readers already announce |
| 162 |
* "image", length is capped by convention, and a description that opens |
| 163 |
* with "an image of" wastes the listener's time. |
| 164 |
* |
| 165 |
* @param string $context Surrounding context, if any. |
| 166 |
* @return string |
| 167 |
*/ |
| 168 |
private function build_prompt(string $context): string { |
| 169 |
$prompt = "Write alt text for this image.\n\n" |
| 170 |
. "Rules:\n" |
| 171 |
. "- One sentence, under 125 characters.\n" |
| 172 |
. "- Describe what is visibly in the image, factually.\n" |
| 173 |
. "- Do NOT start with \"image of\", \"picture of\" or \"photo of\".\n" |
| 174 |
. "- No quotes, no trailing period, no markdown.\n" |
| 175 |
. "- If the image contains meaningful text, include it.\n" |
| 176 |
. "- Reply with the alt text only."; |
| 177 |
|
| 178 |
if ('' !== $context) { |
| 179 |
$prompt .= "\n\nThe image appears in content about: " . $context; |
| 180 |
} |
| 181 |
|
| 182 |
return $prompt; |
| 183 |
} |
| 184 |
|
| 185 |
/** |
| 186 |
* Load an attachment as base64, preferring a scaled size. |
| 187 |
* |
| 188 |
* @param int $attachment_id Attachment ID. |
| 189 |
* @return array{data: string, mime: string}|null |
| 190 |
*/ |
| 191 |
private function read_image(int $attachment_id): ?array { |
| 192 |
$path = $this->resolve_path($attachment_id); |
| 193 |
if (null === $path) { |
| 194 |
return null; |
| 195 |
} |
| 196 |
|
| 197 |
if (!is_readable($path)) { |
| 198 |
return null; |
| 199 |
} |
| 200 |
|
| 201 |
$bytes = filesize($path); |
| 202 |
if (false === $bytes || $bytes > self::MAX_IMAGE_BYTES) { |
| 203 |
return null; |
| 204 |
} |
| 205 |
|
| 206 |
// phpcs:ignore WordPress.WP.AlternativeFunctions.file_get_contents_file_get_contents -- reading a local upload, not a remote fetch. |
| 207 |
$contents = file_get_contents($path); |
| 208 |
if (false === $contents || '' === $contents) { |
| 209 |
return null; |
| 210 |
} |
| 211 |
|
| 212 |
$mime = (string) get_post_mime_type($attachment_id); |
| 213 |
if (!in_array($mime, ['image/jpeg', 'image/png', 'image/gif', 'image/webp'], true)) { |
| 214 |
return null; |
| 215 |
} |
| 216 |
|
| 217 |
// phpcs:ignore WordPress.PHP.DiscouragedPHPFunctions.obfuscation_base64_encode -- the multimodal request bodies of OpenAI, Anthropic and Gemini all carry image bytes as base64. |
| 218 |
return ['data' => base64_encode($contents), 'mime' => $mime]; |
| 219 |
} |
| 220 |
|
| 221 |
/** |
| 222 |
* Absolute path to the smallest adequate version of an attachment. |
| 223 |
* |
| 224 |
* @param int $attachment_id Attachment ID. |
| 225 |
* @return string|null |
| 226 |
*/ |
| 227 |
private function resolve_path(int $attachment_id): ?string { |
| 228 |
$meta = wp_get_attachment_metadata($attachment_id); |
| 229 |
$original = get_attached_file($attachment_id); |
| 230 |
|
| 231 |
if (is_array($meta) && !empty($meta['sizes']) && is_string($original)) { |
| 232 |
$dir = dirname($original); |
| 233 |
foreach (self::PREFERRED_SIZES as $size) { |
| 234 |
if (!empty($meta['sizes'][$size]['file'])) { |
| 235 |
$candidate = $dir . '/' . $meta['sizes'][$size]['file']; |
| 236 |
if (file_exists($candidate)) { |
| 237 |
return $candidate; |
| 238 |
} |
| 239 |
} |
| 240 |
} |
| 241 |
} |
| 242 |
|
| 243 |
return (is_string($original) && file_exists($original)) ? $original : null; |
| 244 |
} |
| 245 |
|
| 246 |
/** |
| 247 |
* OpenAI: typed content parts with a data: URL. |
| 248 |
* |
| 249 |
* @param string $api_key API key. |
| 250 |
* @param string $model Model id. |
| 251 |
* @param string $prompt Instruction. |
| 252 |
* @param array $image ['data' => base64, 'mime' => string]. |
| 253 |
* @return string |
| 254 |
* @throws \Exception On API failure. |
| 255 |
*/ |
| 256 |
private function call_openai(string $api_key, string $model, string $prompt, array $image): string { |
| 257 |
$body = [ |
| 258 |
'model' => $model, |
| 259 |
'messages' => [[ |
| 260 |
'role' => 'user', |
| 261 |
'content' => [ |
| 262 |
['type' => 'text', 'text' => $prompt], |
| 263 |
[ |
| 264 |
'type' => 'image_url', |
| 265 |
'image_url' => ['url' => 'data:' . $image['mime'] . ';base64,' . $image['data']], |
| 266 |
], |
| 267 |
], |
| 268 |
]], |
| 269 |
// Reasoning-safe ceiling, and minimal effort: alt text needs |
| 270 |
// observation, not deliberation (see 6094d9d). |
| 271 |
'max_completion_tokens' => 4000, |
| 272 |
]; |
| 273 |
|
| 274 |
if (0 === strpos($model, 'gpt-5')) { |
| 275 |
$body['reasoning_effort'] = 'minimal'; |
| 276 |
} |
| 277 |
|
| 278 |
$response = $this->post('https://api.openai.com/v1/chat/completions', [ |
| 279 |
'Authorization' => 'Bearer ' . $api_key, |
| 280 |
'Content-Type' => 'application/json', |
| 281 |
], $body); |
| 282 |
|
| 283 |
return (string) ($response['choices'][0]['message']['content'] ?? ''); |
| 284 |
} |
| 285 |
|
| 286 |
/** |
| 287 |
* Anthropic: base64 `source` block, and the API version header. |
| 288 |
* |
| 289 |
* @param string $api_key API key. |
| 290 |
* @param string $model Model id. |
| 291 |
* @param string $prompt Instruction. |
| 292 |
* @param array $image ['data' => base64, 'mime' => string]. |
| 293 |
* @return string |
| 294 |
* @throws \Exception On API failure. |
| 295 |
*/ |
| 296 |
private function call_claude(string $api_key, string $model, string $prompt, array $image): string { |
| 297 |
$response = $this->post('https://api.anthropic.com/v1/messages', [ |
| 298 |
'x-api-key' => $api_key, |
| 299 |
'anthropic-version' => '2023-06-01', |
| 300 |
'Content-Type' => 'application/json', |
| 301 |
], [ |
| 302 |
'model' => $model, |
| 303 |
'max_tokens' => 300, |
| 304 |
'messages' => [[ |
| 305 |
'role' => 'user', |
| 306 |
'content' => [ |
| 307 |
[ |
| 308 |
'type' => 'image', |
| 309 |
'source' => [ |
| 310 |
'type' => 'base64', |
| 311 |
'media_type' => $image['mime'], |
| 312 |
'data' => $image['data'], |
| 313 |
], |
| 314 |
], |
| 315 |
['type' => 'text', 'text' => $prompt], |
| 316 |
], |
| 317 |
]], |
| 318 |
]); |
| 319 |
|
| 320 |
return (string) ($response['content'][0]['text'] ?? ''); |
| 321 |
} |
| 322 |
|
| 323 |
/** |
| 324 |
* Gemini: `inline_data` part, key on the query string. |
| 325 |
* |
| 326 |
* @param string $api_key API key. |
| 327 |
* @param string $model Model id. |
| 328 |
* @param string $prompt Instruction. |
| 329 |
* @param array $image ['data' => base64, 'mime' => string]. |
| 330 |
* @return string |
| 331 |
* @throws \Exception On API failure. |
| 332 |
*/ |
| 333 |
private function call_gemini(string $api_key, string $model, string $prompt, array $image): string { |
| 334 |
$url = sprintf( |
| 335 |
'https://generativelanguage.googleapis.com/v1beta/models/%s:generateContent?key=%s', |
| 336 |
rawurlencode($model), |
| 337 |
rawurlencode($api_key) |
| 338 |
); |
| 339 |
|
| 340 |
$response = $this->post($url, ['Content-Type' => 'application/json'], [ |
| 341 |
'contents' => [[ |
| 342 |
'parts' => [ |
| 343 |
['text' => $prompt], |
| 344 |
['inline_data' => ['mime_type' => $image['mime'], 'data' => $image['data']]], |
| 345 |
], |
| 346 |
]], |
| 347 |
// Gemini 2.5 spends output tokens on reasoning before emitting |
| 348 |
// text, so a tight cap returns an empty candidate. |
| 349 |
'generationConfig' => ['maxOutputTokens' => 2000, 'temperature' => 0.2], |
| 350 |
]); |
| 351 |
|
| 352 |
return (string) ($response['candidates'][0]['content']['parts'][0]['text'] ?? ''); |
| 353 |
} |
| 354 |
|
| 355 |
/** |
| 356 |
* Shared POST + error mapping. |
| 357 |
* |
| 358 |
* @param string $url Endpoint. |
| 359 |
* @param array $headers Headers. |
| 360 |
* @param array $body Payload. |
| 361 |
* @return array Decoded response. |
| 362 |
* @throws \Exception On transport or API error. |
| 363 |
*/ |
| 364 |
private function post(string $url, array $headers, array $body): array { |
| 365 |
$response = wp_remote_post($url, [ |
| 366 |
// Vision calls carry a payload and think for a moment; the 30s |
| 367 |
// default is too tight for a large image on a slow link. |
| 368 |
'timeout' => 60, |
| 369 |
'headers' => $headers, |
| 370 |
'body' => wp_json_encode($body), |
| 371 |
]); |
| 372 |
|
| 373 |
if (is_wp_error($response)) { |
| 374 |
throw new \Exception(esc_html($response->get_error_message())); |
| 375 |
} |
| 376 |
|
| 377 |
$status = (int) wp_remote_retrieve_response_code($response); |
| 378 |
$raw = wp_remote_retrieve_body($response); |
| 379 |
$data = json_decode($raw, true); |
| 380 |
|
| 381 |
if ($status >= 400) { |
| 382 |
$message = $data['error']['message'] ?? ($data['error']['status'] ?? 'Unknown API error'); |
| 383 |
throw new \Exception(sprintf('Vision API error (%d): %s', $status, esc_html((string) $message))); |
| 384 |
} |
| 385 |
|
| 386 |
return is_array($data) ? $data : []; |
| 387 |
} |
| 388 |
|
| 389 |
/** |
| 390 |
* Normalize a model's reply into usable alt text. |
| 391 |
* |
| 392 |
* Models still add wrappers despite instructions, so strip them here rather |
| 393 |
* than storing "Image of a red bicycle." on the attachment forever. |
| 394 |
* |
| 395 |
* @param string $text Raw model output. |
| 396 |
* @return string |
| 397 |
*/ |
| 398 |
private function clean(string $text): string { |
| 399 |
$text = trim(wp_strip_all_tags($text)); |
| 400 |
$text = trim($text, "\"' \t\n\r"); |
| 401 |
$text = (string) preg_replace('/^(an?\s+)?(image|picture|photo|photograph|screenshot)\s+(of|showing|depicting)\s+/i', '', $text); |
| 402 |
$text = (string) preg_replace('/\s+/', ' ', $text); |
| 403 |
$text = rtrim($text, '.'); |
| 404 |
|
| 405 |
if ('' === $text) { |
| 406 |
return ''; |
| 407 |
} |
| 408 |
|
| 409 |
// Alt text longer than ~125 chars is read as noise by screen readers; |
| 410 |
// cut on a word boundary rather than mid-syllable. |
| 411 |
if (mb_strlen($text) > 125) { |
| 412 |
$text = mb_substr($text, 0, 125); |
| 413 |
$space = mb_strrpos($text, ' '); |
| 414 |
if (false !== $space && $space > 60) { |
| 415 |
$text = mb_substr($text, 0, $space); |
| 416 |
} |
| 417 |
} |
| 418 |
|
| 419 |
// ucfirst() is not multibyte-safe: on alt text starting with an |
| 420 |
// accented character ("Éclair…") it corrupts the leading byte. |
| 421 |
return mb_strtoupper(mb_substr($text, 0, 1)) . mb_substr($text, 1); |
| 422 |
} |
| 423 |
} |
| 424 |
|
