@ljoukov/llm 7.0.15 → 7.0.17

package/README.md

@@ -209,9 +209,10 @@ console.log(OPENAI_GPT_IMAGE_2_SIZE_CONSTRAINTS);
 console.log(images[0]?.mimeType, images[0]?.data.byteLength);
 ```
 
-`generateImages()` is typed as a discriminated union by `model`: `gpt-image-2` requests use
-`imageResolution`, while Gemini image requests use `imageSize` (`"1K" | "2K" | "4K"`). For
-`gpt-image-2`, `OPENAI_GPT_IMAGE_2_RESOLUTIONS` exposes the documented popular presets plus
+`generateImages()` is typed as a discriminated union by `model`: `gpt-image-2` and
+`chatgpt-gpt-image-2` requests use `imageResolution`, while Gemini image requests use `imageSize`
+(`"1K" | "2K" | "4K"`). For GPT Image 2, `OPENAI_GPT_IMAGE_2_RESOLUTIONS` exposes the documented
+popular presets plus
 `"auto"`; custom literal `WIDTHxHEIGHT` resolutions are also accepted when they satisfy
 `OPENAI_GPT_IMAGE_2_SIZE_CONSTRAINTS`: each edge must be at most 3840px, each edge must be a
 multiple of 16px, the long edge must be at most 3:1 relative to the short edge, and total pixels
@@ -226,14 +227,20 @@ const images = await generateImages({
   model: "chatgpt-gpt-image-2",
   stylePrompt: "Warm amber desk light, deep blue night, cinematic laboratory mood.",
   imagePrompts: ["A compact lab bench still life with glassware and an open notebook"],
+  imageResolution: "1024x1536",
+  imageQuality: "high",
+  outputFormat: "jpeg",
+  outputCompression: 50,
+  action: "generate",
   numImages: 1,
 });
 ```
 
 That path reuses the same ChatGPT auth setup as other `chatgpt-*` models and sends the request
-through the ChatGPT/Codex Responses `image_generation` built-in tool. It returns PNG images. The
-public Images API controls such as `imageResolution`, `imageQuality`, `outputFormat`, and
-`outputCompression` are intentionally only on the `gpt-image-2` request type.
+through the ChatGPT/Codex Responses `image_generation` built-in tool. `imageResolution`,
+`imageQuality`, `outputFormat`, `outputCompression`, `background`, `moderation`, and `action` are
+passed as tool options. `numImages` is implemented as repeated one-image tool calls because the
+ChatGPT/Codex tool rejects `n` on `tools[0]`.
 
 ### Streaming (response + thoughts + usage)
 

package/dist/index.cjs

@@ -5334,6 +5334,8 @@ function resolveOpenAiReasoningEffort(modelId, thinkingLevel) {
       case "medium":
         return "medium";
       case "high":
+        return "high";
+      case "xhigh":
         return "xhigh";
     }
   }
@@ -5351,7 +5353,7 @@ function toOpenAiReasoningEffort(effort) {
     case "high":
       return "high";
     case "xhigh":
-      return "high";
+      return "xhigh";
   }
 }
 function resolveOpenAiVerbosity(modelId) {
@@ -7644,6 +7646,7 @@ function toGeminiThinkingLevel(thinkingLevel) {
     case "medium":
       return import_genai2.ThinkingLevel.MEDIUM;
     case "high":
+    case "xhigh":
       return import_genai2.ThinkingLevel.HIGH;
   }
 }
@@ -7686,6 +7689,7 @@ function resolveGeminiThinkingBudget(modelId, thinkingLevel) {
       case "medium":
         return 4096;
       case "high":
+      case "xhigh":
         return 32768;
     }
   }
@@ -7696,6 +7700,7 @@ function resolveGeminiThinkingBudget(modelId, thinkingLevel) {
       case "medium":
         return 8192;
       case "high":
+      case "xhigh":
         return 24576;
     }
   }
@@ -7706,6 +7711,7 @@ function resolveGeminiThinkingBudget(modelId, thinkingLevel) {
       case "medium":
         return 8192;
       case "high":
+      case "xhigh":
         return 16384;
     }
   }
@@ -10578,7 +10584,7 @@ function buildOpenAiImagePrompt(params) {
     params.imagePrompt.trim()
   ].filter((line) => line.length > 0).join("\n");
 }
-function resolveOpenAiImageRequestParams(request) {
+function resolveGptImage2RequestParams(request) {
   if (request.partialImages !== void 0) {
     throw new Error("partialImages is only supported for streaming image generation.");
   }
@@ -10592,7 +10598,7 @@ function resolveOpenAiImageRequestParams(request) {
   const sizeValidation = validateOpenAiGptImage2Resolution(size);
   if (!sizeValidation.valid) {
     throw new Error(
-      `imageResolution ${JSON.stringify(size)} is not supported by gpt-image-2: ${sizeValidation.reason}`
+      `imageResolution ${JSON.stringify(size)} is not supported by ${request.model}: ${sizeValidation.reason}`
     );
   }
   return {
@@ -10635,7 +10641,7 @@ async function generateImagesWithOpenAiImageApi(request) {
     model: request.model
   });
   const startedAtMs = Date.now();
-  const params = resolveOpenAiImageRequestParams(request);
+  const params = resolveGptImage2RequestParams(request);
   const styleImages = await createOpenAiStyleImageFiles(request.styleImages);
   const hasStyleImages = Boolean(styleImages && styleImages.length > 0);
   const outputMimeType = resolveOpenAiImageMimeType(params.outputFormat);
@@ -10763,7 +10769,8 @@ async function generateImagesWithChatGptImageTool(request) {
     model: request.model
   });
   const startedAtMs = Date.now();
-  const numImagesPerPrompt = request.numImages ?? 1;
+  const params = resolveGptImage2RequestParams(request);
+  const outputMimeType = resolveOpenAiImageMimeType(params.outputFormat);
   let totalUsage;
   let costUsd = 0;
   let outputImages = 0;
@@ -10771,7 +10778,7 @@ async function generateImagesWithChatGptImageTool(request) {
     type: "llm.call.started",
     imagePromptCount: promptEntries.length,
     styleImageCount: request.styleImages?.length ?? 0,
-    numImagesPerPrompt
+    numImagesPerPrompt: params.n
   });
   try {
     const images = [];
@@ -10781,7 +10788,7 @@ async function generateImagesWithChatGptImageTool(request) {
         imagePrompt,
         hasStyleImages: Boolean(request.styleImages && request.styleImages.length > 0)
       });
-      for (let imageIndex = 0; imageIndex < numImagesPerPrompt; imageIndex += 1) {
+      for (let imageIndex = 0; imageIndex < params.n; imageIndex += 1) {
         const chatGptInput = toChatGptInput(
           buildChatGptImageInputContent({
             prompt,
@@ -10798,11 +10805,22 @@ async function generateImagesWithChatGptImageTool(request) {
             model: providerInfo.model,
             store: false,
             stream: true,
-            instructions: chatGptInput.instructions ?? "Use the image_generation tool to generate exactly one PNG image. Do not return prose instead of the image.",
+            instructions: chatGptInput.instructions ?? "Use the image_generation tool to generate exactly one image. Do not return prose instead of the image.",
             input: preparedInput,
             tool_choice: "required",
             parallel_tool_calls: false,
-            tools: [{ type: "image_generation", output_format: "png" }]
+            tools: [
+              {
+                type: "image_generation",
+                size: params.size,
+                quality: params.quality,
+                output_format: params.outputFormat ?? "png",
+                ...request.outputCompression !== void 0 ? { output_compression: request.outputCompression } : {},
+                ...params.background ? { background: params.background } : {},
+                ...params.moderation ? { moderation: params.moderation } : {},
+                ...request.action ? { action: request.action } : {}
+              }
+            ]
           },
           signal: request.signal
         });
@@ -10814,7 +10832,7 @@ async function generateImagesWithChatGptImageTool(request) {
         }
         for (const call of result.imageGenerationCalls) {
           images.push({
-            mimeType: "image/png",
+            mimeType: outputMimeType,
             data: import_node_buffer4.Buffer.from(call.result, "base64")
           });
         }
@@ -10825,8 +10843,8 @@ async function generateImagesWithChatGptImageTool(request) {
           modelId: request.model,
           tokens: usage,
           responseImages: result.imageGenerationCalls.length,
-          imageSize: "1024x1024",
-          imageQuality: "medium"
+          imageSize: params.size,
+          imageQuality: params.quality
         });
       }
     }
@@ -10838,7 +10856,7 @@ async function generateImagesWithChatGptImageTool(request) {
       usage: totalUsage,
       costUsd,
       imageCount: images.length,
-      attempts: promptEntries.length * numImagesPerPrompt
+      attempts: promptEntries.length * params.n
     });
     return images;
   } catch (error) {