visual-ai-assertions 0.16.0 → 0.19.0

package/dist/index.js

@@ -5,6 +5,13 @@ var ReasoningEffort = {
   HIGH: "high",
   XHIGH: "xhigh"
 };
+var ImageDetail = {
+  AUTO: "auto",
+  LOW: "low",
+  HIGH: "high"
+};
+var DEFAULT_IMAGE_DETAIL = ImageDetail.AUTO;
+var DEFAULT_MAX_IMAGE_DIMENSION = 1568;
 var Provider = {
   ANTHROPIC: "anthropic",
   OPENAI: "openai",
@@ -14,6 +21,7 @@ var Provider = {
 var Model = {
   Anthropic: {
     FABLE_5: "claude-fable-5",
+    OPUS_5: "claude-opus-5",
     OPUS_4_8: "claude-opus-4-8",
     OPUS_4_7: "claude-opus-4-7",
     OPUS_4_6: "claude-opus-4-6",
@@ -34,6 +42,8 @@ var Model = {
     GPT_5_MINI: "gpt-5-mini"
   },
   Google: {
+    GEMINI_3_8_FLASH: "gemini-3.8-flash",
+    GEMINI_3_7_FLASH: "gemini-3.7-flash",
     GEMINI_3_6_FLASH: "gemini-3.6-flash",
     GEMINI_3_5_FLASH: "gemini-3.5-flash",
     GEMINI_3_5_FLASH_LITE: "gemini-3.5-flash-lite",
@@ -47,9 +57,11 @@ var Model = {
    * recognizes them. All listed models accept image input.
    */
   OpenRouter: {
+    GROK_4_6: "x-ai/grok-4.6",
     GROK_4_5: "x-ai/grok-4.5",
     KIMI_K3: "moonshotai/kimi-k3",
     KIMI_K2_7_CODE: "moonshotai/kimi-k2.7-code",
+    QWEN_3_8_MAX: "qwen/qwen3.8-max",
     QWEN_3_7_PLUS: "qwen/qwen3.7-plus",
     QWEN_3_6_FLASH: "qwen/qwen3.6-flash"
   }
@@ -532,6 +544,7 @@ function parseRetryAfter(value) {
 // src/providers/anthropic.ts
 var XHIGH_CAPABLE_MODELS = /* @__PURE__ */ new Set([
   Model.Anthropic.FABLE_5,
+  Model.Anthropic.OPUS_5,
   Model.Anthropic.OPUS_4_8,
   Model.Anthropic.OPUS_4_7,
   Model.Anthropic.SONNET_5
@@ -627,7 +640,10 @@ var AnthropicDriver = class {
         text,
         usage: {
           inputTokens: message.usage.input_tokens,
-          outputTokens: message.usage.output_tokens
+          outputTokens: message.usage.output_tokens,
+          ...message.usage.cache_read_input_tokens !== void 0 && {
+            cachedInputTokens: message.usage.cache_read_input_tokens
+          }
         }
       };
     } catch (err) {
@@ -644,23 +660,41 @@ function needsCodeExecution(model) {
   return match !== null && match[1] !== void 0 && parseInt(match[1], 10) >= 3;
 }
 var GOOGLE_THINKING_LEVEL = {
-  low: "minimal",
-  medium: "low",
-  high: "medium",
+  low: "low",
+  medium: "medium",
+  high: "high",
   xhigh: "high"
 };
+var GOOGLE_MEDIA_RESOLUTION = {
+  low: "MEDIA_RESOLUTION_LOW",
+  high: "MEDIA_RESOLUTION_HIGH"
+};
+function toGeminiUsage(um) {
+  if (!um) return void 0;
+  const thoughts = um.thoughtsTokenCount ?? 0;
+  return {
+    inputTokens: um.promptTokenCount ?? 0,
+    outputTokens: (um.candidatesTokenCount ?? 0) + thoughts,
+    ...um.thoughtsTokenCount !== void 0 && { reasoningTokens: um.thoughtsTokenCount },
+    ...um.cachedContentTokenCount !== void 0 && {
+      cachedInputTokens: um.cachedContentTokenCount
+    }
+  };
+}
 var GoogleDriver = class {
   client;
   model;
   maxTokens;
   apiKeyOrEnv;
   reasoningEffort;
+  imageDetail;
   constructor(config) {
     this.model = config.model;
     this.maxTokens = config.maxTokens;
     this.client = null;
     this.apiKeyOrEnv = config.apiKey;
     this.reasoningEffort = config.reasoningEffort;
+    this.imageDetail = config.imageDetail;
   }
   toGeminiParts(images) {
     return images.map((img) => ({
@@ -700,6 +734,9 @@ var GoogleDriver = class {
             thinkingConfig: {
               thinkingLevel: GOOGLE_THINKING_LEVEL[this.reasoningEffort]
             }
+          },
+          ...this.imageDetail && GOOGLE_MEDIA_RESOLUTION[this.imageDetail] && {
+            mediaResolution: GOOGLE_MEDIA_RESOLUTION[this.imageDetail]
           }
         }
       });
@@ -717,14 +754,9 @@ var GoogleDriver = class {
         );
       }
       const text = response.text ?? "";
-      const thoughtsTokenCount = response.usageMetadata?.thoughtsTokenCount;
       return {
         text,
-        usage: response.usageMetadata ? {
-          inputTokens: response.usageMetadata.promptTokenCount ?? 0,
-          outputTokens: response.usageMetadata.candidatesTokenCount ?? 0,
-          ...thoughtsTokenCount !== void 0 && { reasoningTokens: thoughtsTokenCount }
-        } : void 0
+        usage: toGeminiUsage(response.usageMetadata)
       };
     } catch (err) {
       if (err instanceof VisualAITruncationError || err instanceof VisualAIProviderError) throw err;
@@ -755,10 +787,7 @@ var GoogleDriver = class {
       return {
         imageData: Buffer.from(imagePart.inlineData.data, "base64"),
         mimeType: imagePart.inlineData.mimeType,
-        usage: response.usageMetadata ? {
-          inputTokens: response.usageMetadata.promptTokenCount ?? 0,
-          outputTokens: response.usageMetadata.candidatesTokenCount ?? 0
-        } : void 0
+        usage: toGeminiUsage(response.usageMetadata)
       };
     } catch (err) {
       if (err instanceof VisualAIProviderError) throw err;
@@ -774,12 +803,14 @@ var OpenAIDriver = class {
   maxTokens;
   apiKeyOrEnv;
   reasoningEffort;
+  imageDetail;
   constructor(config) {
     this.model = config.model;
     this.maxTokens = config.maxTokens;
     this.client = null;
     this.apiKeyOrEnv = config.apiKey;
     this.reasoningEffort = config.reasoningEffort;
+    this.imageDetail = config.imageDetail;
   }
   async getClient() {
     if (this.client) return this.client;
@@ -801,9 +832,11 @@ var OpenAIDriver = class {
   }
   async sendMessage(images, prompt, options) {
     const client = await this.getClient();
+    const detail = this.imageDetail && this.imageDetail !== "auto" ? this.imageDetail : void 0;
     const imageBlocks = images.map((img) => ({
       type: "input_image",
-      image_url: `data:${img.mimeType};base64,${img.base64}`
+      image_url: `data:${img.mimeType};base64,${img.base64}`,
+      ...detail ? { detail } : {}
     }));
     try {
       const format = options?.responseSchema ? {
@@ -828,21 +861,23 @@ var OpenAIDriver = class {
       }
       const response = await client.responses.create(requestParams);
       if (response.status && response.status !== "completed") {
-        const detail = response.incomplete_details?.reason ? ` (${response.incomplete_details.reason})` : "";
+        const detail2 = response.incomplete_details?.reason ? ` (${response.incomplete_details.reason})` : "";
         throw new VisualAITruncationError(
-          `Response truncated: OpenAI returned status "${response.status}"${detail}. The model exhausted the output token budget (${this.maxTokens} tokens). This commonly happens with higher reasoning effort levels. Increase maxTokens in your config (e.g., maxTokens: 16384) or lower reasoningEffort.`,
+          `Response truncated: OpenAI returned status "${response.status}"${detail2}. The model exhausted the output token budget (${this.maxTokens} tokens). This commonly happens with higher reasoning effort levels. Increase maxTokens in your config (e.g., maxTokens: 16384) or lower reasoningEffort.`,
           response.output_text ?? "",
           this.maxTokens
         );
       }
       const text = response.output_text ?? "";
       const reasoningTokens = response.usage?.output_tokens_details?.reasoning_tokens;
+      const cachedInputTokens = response.usage?.input_tokens_details?.cached_tokens;
       return {
         text,
         usage: response.usage ? {
           inputTokens: response.usage.input_tokens,
           outputTokens: response.usage.output_tokens,
-          ...reasoningTokens !== void 0 && { reasoningTokens }
+          ...reasoningTokens !== void 0 && { reasoningTokens },
+          ...cachedInputTokens !== void 0 && { cachedInputTokens }
         } : void 0
       };
     } catch (err) {
@@ -866,12 +901,14 @@ var OpenRouterDriver = class {
   maxTokens;
   apiKeyOrEnv;
   reasoningEffort;
+  imageDetail;
   constructor(config) {
     this.model = config.model;
     this.maxTokens = config.maxTokens;
     this.client = null;
     this.apiKeyOrEnv = config.apiKey;
     this.reasoningEffort = config.reasoningEffort;
+    this.imageDetail = config.imageDetail;
   }
   async getClient() {
     if (this.client) return this.client;
@@ -895,9 +932,13 @@ var OpenRouterDriver = class {
   }
   async sendMessage(images, prompt, options) {
     const client = await this.getClient();
+    const detail = this.imageDetail && this.imageDetail !== "auto" ? this.imageDetail : void 0;
     const imageParts = images.map((img) => ({
       type: "image_url",
-      image_url: { url: `data:${img.mimeType};base64,${img.base64}` }
+      image_url: {
+        url: `data:${img.mimeType};base64,${img.base64}`,
+        ...detail ? { detail } : {}
+      }
     }));
     try {
       const responseFormat = options?.responseSchema ? {
@@ -938,12 +979,16 @@ var OpenRouterDriver = class {
         );
       }
       const reasoningTokens = response.usage?.completion_tokens_details?.reasoning_tokens;
+      const cachedInputTokens = response.usage?.prompt_tokens_details?.cached_tokens;
+      const cost = response.usage?.cost;
       return {
         text,
         usage: response.usage ? {
           inputTokens: response.usage.prompt_tokens,
           outputTokens: response.usage.completion_tokens,
-          ...reasoningTokens !== void 0 && { reasoningTokens }
+          ...reasoningTokens !== void 0 && { reasoningTokens },
+          ...cachedInputTokens !== void 0 && { cachedInputTokens },
+          ...cost !== void 0 && { cost }
         } : void 0
       };
     } catch (err) {
@@ -1029,6 +1074,8 @@ function resolveConfig(config) {
     model,
     maxTokens,
     reasoningEffort: config.reasoningEffort,
+    maxImageDimension: config.maxImageDimension ?? DEFAULT_MAX_IMAGE_DIMENSION,
+    imageDetail: config.imageDetail ?? DEFAULT_IMAGE_DETAIL,
     debug,
     debugPrompt,
     debugResponse,
@@ -1043,6 +1090,10 @@ var PRICING_TABLE = {
     inputPricePerToken: 10 / PER_MILLION,
     outputPricePerToken: 50 / PER_MILLION
   },
+  [`${Provider.ANTHROPIC}:${Model.Anthropic.OPUS_5}`]: {
+    inputPricePerToken: 5 / PER_MILLION,
+    outputPricePerToken: 25 / PER_MILLION
+  },
   [`${Provider.ANTHROPIC}:${Model.Anthropic.OPUS_4_8}`]: {
     inputPricePerToken: 5 / PER_MILLION,
     outputPricePerToken: 25 / PER_MILLION
@@ -1072,12 +1123,12 @@ var PRICING_TABLE = {
     outputPricePerToken: 30 / PER_MILLION
   },
   [`${Provider.OPENAI}:${Model.OpenAI.GPT_5_6_TERRA}`]: {
-    inputPricePerToken: 2.5 / PER_MILLION,
-    outputPricePerToken: 15 / PER_MILLION
+    inputPricePerToken: 2 / PER_MILLION,
+    outputPricePerToken: 12 / PER_MILLION
   },
   [`${Provider.OPENAI}:${Model.OpenAI.GPT_5_6_LUNA}`]: {
-    inputPricePerToken: 1 / PER_MILLION,
-    outputPricePerToken: 6 / PER_MILLION
+    inputPricePerToken: 0.2 / PER_MILLION,
+    outputPricePerToken: 1.2 / PER_MILLION
   },
   [`${Provider.OPENAI}:${Model.OpenAI.GPT_5_5}`]: {
     inputPricePerToken: 5 / PER_MILLION,
@@ -1107,6 +1158,18 @@ var PRICING_TABLE = {
     inputPricePerToken: 0.25 / PER_MILLION,
     outputPricePerToken: 2 / PER_MILLION
   },
+  // Introductory pricing through 2026-12-31; reverts to $1.50/$7.50 per MTok
+  // on 2027-01-01 (https://blog.google/.../3-8-flash-and-3-8-flash-cyber/).
+  [`${Provider.GOOGLE}:${Model.Google.GEMINI_3_8_FLASH}`]: {
+    inputPricePerToken: 0.75 / PER_MILLION,
+    outputPricePerToken: 3.75 / PER_MILLION
+  },
+  // Introductory pricing through 2026-12-31; reverts to $1.50/$7.50 per MTok
+  // on 2027-01-01 (https://blog.google/.../introducing-gemini-3-7-flash/).
+  [`${Provider.GOOGLE}:${Model.Google.GEMINI_3_7_FLASH}`]: {
+    inputPricePerToken: 0.75 / PER_MILLION,
+    outputPricePerToken: 3.75 / PER_MILLION
+  },
   [`${Provider.GOOGLE}:${Model.Google.GEMINI_3_6_FLASH}`]: {
     inputPricePerToken: 1.5 / PER_MILLION,
     outputPricePerToken: 7.5 / PER_MILLION
@@ -1133,6 +1196,10 @@ var PRICING_TABLE = {
   },
   // OpenRouter passes through upstream per-model pricing (verified 2026-07-22
   // against https://openrouter.ai/api/v1/models).
+  [`${Provider.OPENROUTER}:${Model.OpenRouter.GROK_4_6}`]: {
+    inputPricePerToken: 2 / PER_MILLION,
+    outputPricePerToken: 6 / PER_MILLION
+  },
   [`${Provider.OPENROUTER}:${Model.OpenRouter.GROK_4_5}`]: {
     inputPricePerToken: 2 / PER_MILLION,
     outputPricePerToken: 6 / PER_MILLION
@@ -1145,6 +1212,10 @@ var PRICING_TABLE = {
     inputPricePerToken: 0.82 / PER_MILLION,
     outputPricePerToken: 3.75 / PER_MILLION
   },
+  [`${Provider.OPENROUTER}:${Model.OpenRouter.QWEN_3_8_MAX}`]: {
+    inputPricePerToken: 2 / PER_MILLION,
+    outputPricePerToken: 6 / PER_MILLION
+  },
   [`${Provider.OPENROUTER}:${Model.OpenRouter.QWEN_3_7_PLUS}`]: {
     inputPricePerToken: 0.32 / PER_MILLION,
     outputPricePerToken: 1.28 / PER_MILLION
@@ -1174,8 +1245,9 @@ function usageLog(config, method, usage) {
   const costStr = usage.estimatedCost !== void 0 ? `$${usage.estimatedCost.toFixed(6)}` : "unknown";
   const reasoningStr = config.reasoningEffort ? `reasoning: ${config.reasoningEffort}` : `reasoning: ${PROVIDER_DEFAULT_REASONING[config.provider]} (provider default)`;
   const reasoningTokenStr = usage.reasoningTokens !== void 0 ? ` (${usage.reasoningTokens} reasoning)` : "";
+  const cachedTokenStr = usage.cachedInputTokens !== void 0 ? ` (${usage.cachedInputTokens} cached)` : "";
   process.stderr.write(
-    `[visual-ai-assertions] ${method} usage: ${usage.inputTokens} input + ${usage.outputTokens} output${reasoningTokenStr} tokens (${costStr}) in ${usage.durationSeconds?.toFixed(3) ?? "0.000"}s [${config.model}, ${reasoningStr}]
+    `[visual-ai-assertions] ${method} usage: ${usage.inputTokens} input${cachedTokenStr} + ${usage.outputTokens} output${reasoningTokenStr} tokens (${costStr}) in ${usage.durationSeconds?.toFixed(3) ?? "0.000"}s [${config.model}, ${reasoningStr}]
 `
   );
 }
@@ -1186,7 +1258,11 @@ function processUsage(method, rawUsage, durationSeconds, config) {
     inputTokens,
     outputTokens,
     ...rawUsage?.reasoningTokens !== void 0 && { reasoningTokens: rawUsage.reasoningTokens },
+    ...rawUsage?.cachedInputTokens !== void 0 && {
+      cachedInputTokens: rawUsage.cachedInputTokens
+    },
     estimatedCost: calculateCost(config.provider, config.model, inputTokens, outputTokens),
+    ...rawUsage?.cost !== void 0 && { reportedCost: rawUsage.cost },
     durationSeconds
   };
   usageLog(config, method, usage);
@@ -1230,7 +1306,9 @@ import sharp from "sharp";
 var DIFF_ALLOWED_MODELS = /* @__PURE__ */ new Set([
   Model.Google.GEMINI_3_FLASH_PREVIEW,
   Model.Google.GEMINI_3_5_FLASH,
-  Model.Google.GEMINI_3_6_FLASH
+  Model.Google.GEMINI_3_6_FLASH,
+  Model.Google.GEMINI_3_7_FLASH,
+  Model.Google.GEMINI_3_8_FLASH
 ]);
 async function generateAiDiff(imgA, imgB, model, driver) {
   if (!driver.generateImage) {
@@ -1320,7 +1398,6 @@ var EXTENSION_TO_MIME = {
   ".webp": "image/webp",
   ".gif": "image/gif"
 };
-var MAX_DIMENSION = 1568;
 var URL_FETCH_TIMEOUT_MS = 1e4;
 function isSupportedMimeType(value) {
   return SUPPORTED_FORMATS.has(value);
@@ -1344,14 +1421,14 @@ function detectMimeType(data) {
   }
   throw new VisualAIImageError("Unable to detect image format from file content");
 }
-async function resizeIfNeeded(data, mimeType) {
+async function resizeIfNeeded(data, mimeType, maxDimension) {
   if (mimeType === "image/gif") {
     return data;
   }
   if (mimeType === "image/png" && data.length >= 24) {
     const width2 = data.readUInt32BE(16);
     const height2 = data.readUInt32BE(20);
-    if (width2 <= MAX_DIMENSION && height2 <= MAX_DIMENSION) {
+    if (width2 <= maxDimension && height2 <= maxDimension) {
       return data;
     }
   }
@@ -1359,12 +1436,12 @@ async function resizeIfNeeded(data, mimeType) {
   const metadata = await pipeline.metadata();
   const width = metadata.width ?? 0;
   const height = metadata.height ?? 0;
-  if (width <= MAX_DIMENSION && height <= MAX_DIMENSION) {
+  if (width <= maxDimension && height <= maxDimension) {
     return data;
   }
   return pipeline.resize({
-    width: MAX_DIMENSION,
-    height: MAX_DIMENSION,
+    width: maxDimension,
+    height: maxDimension,
     fit: "inside",
     withoutEnlargement: true
   }).toBuffer();
@@ -1419,7 +1496,7 @@ function loadFromBase64(input) {
   }
   return { data, mimeType: mimeType ?? detectMimeType(data) };
 }
-async function normalizeImage(input) {
+async function normalizeImage(input, maxDimension = DEFAULT_MAX_IMAGE_DIMENSION) {
   let data;
   let mimeType;
   if (Buffer.isBuffer(input)) {
@@ -1448,7 +1525,7 @@ async function normalizeImage(input) {
       "Invalid image input: expected Buffer, Uint8Array, file path, URL, or base64 string"
     );
   }
-  data = await resizeIfNeeded(data, mimeType);
+  data = await resizeIfNeeded(data, mimeType, maxDimension);
   let cachedBase64;
   return {
     data,
@@ -1752,7 +1829,7 @@ async function probeDurationSeconds(videoPath) {
     });
   });
 }
-async function extractFrames(videoPath, options = {}) {
+async function extractFrames(videoPath, options = {}, maxDimension = FRAME_MAX_DIMENSION) {
   const fps = options.fps ?? DEFAULT_FPS;
   const maxFrames = options.maxFrames ?? DEFAULT_MAX_FRAMES;
   const maxDurationSeconds = options.maxDurationSeconds ?? DEFAULT_MAX_DURATION_SECONDS;
@@ -1781,7 +1858,7 @@ async function extractFrames(videoPath, options = {}) {
   }
   const outputDir = await mkdtemp(join2(tmpdir(), "visual-ai-frames-"));
   try {
-    const filter = `fps=${fps},scale='if(gt(iw,ih),min(${FRAME_MAX_DIMENSION},iw),-2)':'if(gt(iw,ih),-2,min(${FRAME_MAX_DIMENSION},ih))':flags=area`;
+    const filter = `fps=${fps},scale='if(gt(iw,ih),min(${maxDimension},iw),-2)':'if(gt(iw,ih),-2,min(${maxDimension},ih))':flags=area`;
     await new Promise((resolve2, reject) => {
       let settled = false;
       const cmd = ffmpeg(videoPath);
@@ -1879,7 +1956,7 @@ function isFramesInput(input) {
 function isTimestampedFrameInput(frame) {
   return typeof frame === "object" && !Buffer.isBuffer(frame) && !(frame instanceof Uint8Array) && "image" in frame;
 }
-async function normalizeFrames(input) {
+async function normalizeFrames(input, maxDimension = DEFAULT_MAX_IMAGE_DIMENSION) {
   const rawFrames = input.frames;
   const fps = input.fps ?? DEFAULT_FPS;
   if (rawFrames.length === 0) {
@@ -1904,7 +1981,7 @@ async function normalizeFrames(input) {
           `Invalid timestampSeconds for frame ${index}: ${String(timestampSeconds)}. Must be a finite number >= 0.`
         );
       }
-      const image = await normalizeImage(imageInput);
+      const image = await normalizeImage(imageInput, maxDimension);
       return {
         data: image.data,
         mimeType: image.mimeType,
@@ -1920,14 +1997,14 @@ async function normalizeFrames(input) {
   await saveDebugFrames(frames);
   return { kind: "video", frames, durationSeconds };
 }
-async function normalizeMedia(input, videoOptions) {
+async function normalizeMedia(input, videoOptions, maxDimension = DEFAULT_MAX_IMAGE_DIMENSION) {
   if (isFramesInput(input)) {
-    return normalizeFrames(input);
+    return normalizeFrames(input, maxDimension);
   }
   if (isVideoInput(input)) {
     const { path, cleanup } = await resolveVideoToPath(input);
     try {
-      const { frames, durationSeconds } = await extractFrames(path, videoOptions);
+      const { frames, durationSeconds } = await extractFrames(path, videoOptions, maxDimension);
       await saveDebugFrames(frames);
       return { kind: "video", frames, durationSeconds };
     } finally {
@@ -1937,7 +2014,7 @@ async function normalizeMedia(input, videoOptions) {
       }
     }
   }
-  const image = await normalizeImage(input);
+  const image = await normalizeImage(input, maxDimension);
   return { kind: "image", image };
 }
 
@@ -1978,7 +2055,17 @@ var UsageInfoSchema = z.object({
   outputTokens: z.number(),
   /** Reasoning/thinking tokens consumed by the model (informational, typically included within outputTokens). */
   reasoningTokens: z.number().optional(),
+  /**
+   * Prompt tokens served from the provider's cache, when reported. Informational
+   * only — `estimatedCost` does not apply a cache discount, because providers
+   * differ on whether these are counted inside `inputTokens` (OpenAI, OpenRouter,
+   * Google) or billed as a separate bucket alongside it (Anthropic).
+   */
+  cachedInputTokens: z.number().optional(),
+  /** Cost in USD from the library's local pricing table (inputTokens/outputTokens × per-model rates). */
   estimatedCost: z.number().optional(),
+  /** Actual cost in USD reported by the provider itself, when available (OpenRouter). Authoritative over `estimatedCost`. */
+  reportedCost: z.number().optional(),
   durationSeconds: z.number().nonnegative().optional()
 });
 var BaseResultSchema = z.object({
@@ -2118,16 +2205,18 @@ function visualAI(config = {}) {
     apiKey: resolvedConfig.apiKey,
     model: resolvedConfig.model,
     maxTokens: resolvedConfig.maxTokens,
-    reasoningEffort: resolvedConfig.reasoningEffort
+    reasoningEffort: resolvedConfig.reasoningEffort,
+    imageDetail: resolvedConfig.imageDetail
   };
   const driver = createDriver(resolvedConfig.provider, driverConfig);
+  const maxImageDimension = resolvedConfig.maxImageDimension;
   async function checkElementsVisibility(image, elements, visible, options) {
     const methodName = visible ? "elementsVisible" : "elementsHidden";
     if (elements.length === 0) {
       throw new VisualAIConfigError(`At least one element is required for ${methodName}()`);
     }
     return withErrorDebug(resolvedConfig, methodName, async () => {
-      const img = await normalizeImage(image);
+      const img = await normalizeImage(image, maxImageDimension);
       const prompt = buildElementsVisibilityPrompt(elements, visible, options);
       debugLog(resolvedConfig, `${methodName} prompt`, prompt, "prompt");
       const response = await timedSendMessage(driver, [img], prompt, checkSchemaOptions);
@@ -2146,7 +2235,7 @@ function visualAI(config = {}) {
         throw new VisualAIConfigError("At least one statement is required for check()");
       }
       return withErrorDebug(resolvedConfig, "check", async () => {
-        const media = await normalizeMedia(input, options?.video);
+        const media = await normalizeMedia(input, options?.video, maxImageDimension);
         const { images, mediaContext, framesMetadata } = mediaToProviderInputs(media);
         const prompt = buildCheckPrompt(stmts, {
           instructions: options?.instructions,
@@ -2165,7 +2254,7 @@ function visualAI(config = {}) {
     },
     async ask(input, userPrompt, options) {
       return withErrorDebug(resolvedConfig, "ask", async () => {
-        const media = await normalizeMedia(input, options?.video);
+        const media = await normalizeMedia(input, options?.video, maxImageDimension);
         const { images, mediaContext, framesMetadata } = mediaToProviderInputs(media);
         const prompt = buildAskPrompt(userPrompt, {
           instructions: options?.instructions,
@@ -2184,7 +2273,10 @@ function visualAI(config = {}) {
     },
     async compare(imageA, imageB, options) {
       return withErrorDebug(resolvedConfig, "compare", async () => {
-        const [imgA, imgB] = await Promise.all([normalizeImage(imageA), normalizeImage(imageB)]);
+        const [imgA, imgB] = await Promise.all([
+          normalizeImage(imageA, maxImageDimension),
+          normalizeImage(imageB, maxImageDimension)
+        ]);
         const prompt = buildComparePrompt({
           userPrompt: options?.prompt,
           instructions: options?.instructions
@@ -2222,7 +2314,7 @@ function visualAI(config = {}) {
     },
     async accessibility(image, options) {
       return withErrorDebug(resolvedConfig, "accessibility", async () => {
-        const img = await normalizeImage(image);
+        const img = await normalizeImage(image, maxImageDimension);
         const prompt = buildAccessibilityPrompt(options);
         debugLog(resolvedConfig, "accessibility prompt", prompt, "prompt");
         const response = await timedSendMessage(driver, [img], prompt, checkSchemaOptions);
@@ -2241,7 +2333,7 @@ function visualAI(config = {}) {
     },
     async layout(image, options) {
       return withErrorDebug(resolvedConfig, "layout", async () => {
-        const img = await normalizeImage(image);
+        const img = await normalizeImage(image, maxImageDimension);
         const prompt = buildLayoutPrompt(options);
         debugLog(resolvedConfig, "layout prompt", prompt, "prompt");
         const response = await timedSendMessage(driver, [img], prompt, checkSchemaOptions);
@@ -2255,7 +2347,7 @@ function visualAI(config = {}) {
     },
     async pageLoad(image, options) {
       return withErrorDebug(resolvedConfig, "pageLoad", async () => {
-        const img = await normalizeImage(image);
+        const img = await normalizeImage(image, maxImageDimension);
         const prompt = buildPageLoadPrompt(options);
         debugLog(resolvedConfig, "pageLoad prompt", prompt, "prompt");
         const response = await timedSendMessage(driver, [img], prompt, checkSchemaOptions);
@@ -2269,7 +2361,7 @@ function visualAI(config = {}) {
     },
     async content(image, options) {
       return withErrorDebug(resolvedConfig, "content", async () => {
-        const img = await normalizeImage(image);
+        const img = await normalizeImage(image, maxImageDimension);
         const prompt = buildContentPrompt(options);
         debugLog(resolvedConfig, "content prompt", prompt, "prompt");
         const response = await timedSendMessage(driver, [img], prompt, checkSchemaOptions);
@@ -2347,6 +2439,7 @@ export {
   ConfidenceSchema,
   Content,
   DEFAULT_MODELS,
+  ImageDetail,
   IssueCategorySchema,
   IssuePrioritySchema,
   IssueSchema,