@shop-os/modelkit 0.1.3 → 0.1.4

package/dist/index.d.cts

@@ -12,7 +12,7 @@ import { z } from 'zod';
 /** The AI SDK video model spelling: a catalog id or a provider video model instance. */
 type VideoModel = Parameters<typeof experimental_generateVideo>[0]["model"];
 type AspectRatio = "1:1" | "16:9" | "9:16" | "4:3" | "3:4" | "3:2" | "2:3" | "4:5" | "5:4" | "21:9" | "9:21" | "97:30" | "adaptive";
-type ImageSize = "1K" | "2K" | "4K";
+type ImageSize = "512" | "1K" | "2K" | "4K";
 type VideoResolution = "360p" | "480p" | "720p" | "768p" | "1080p" | "2k" | "4k";
 /** "xhigh" and "max" exist only on gpt-image-2.5; gpt-image-2 rejects them at the provider. */
 type ImageQuality = "auto" | "low" | "medium" | "high" | "xhigh" | "max";
@@ -37,6 +37,8 @@ interface ImageRequest {
     seed?: number;
     count?: number;
     outputFormat?: OutputFormat;
+    /** Upscale factor for upscaler rows; adapters map it to the provider's field. */
+    scaleFactor?: number;
 }
 /**
  * What a caller passes to generateImage. Same field names as the AI SDK's
@@ -120,7 +122,7 @@ interface MediaPart {
     cleanup?: () => Promise<void>;
 }
 type RequiredInput = "firstFrame" | "sourceVideo" | "references";
-type SupportedOption = "seed" | "quality" | "negativePrompt" | "cfgScale" | "generateAudio" | "promptExpansion" | "lastFrame" | "count" | "outputFormat" | "instrumental";
+type SupportedOption = "seed" | "quality" | "negativePrompt" | "cfgScale" | "generateAudio" | "promptExpansion" | "lastFrame" | "count" | "outputFormat" | "scaleFactor" | "instrumental";
 interface ReferenceLimits {
     images?: number;
     videos?: number;
@@ -208,6 +210,8 @@ interface ImageResult {
     images: ImageBytes[];
     usage: UsageEvent;
     raw?: unknown;
+    /** The provider URL the images were downloaded from, when the provider hands one back. */
+    outputUrl?: string;
 }
 /** What generateImage returns: the adapter result plus `image`, SDK accessors and warnings. */
 interface ImageCallResult extends ImageResult {

package/dist/index.d.ts

@@ -12,7 +12,7 @@ import { z } from 'zod';
 /** The AI SDK video model spelling: a catalog id or a provider video model instance. */
 type VideoModel = Parameters<typeof experimental_generateVideo>[0]["model"];
 type AspectRatio = "1:1" | "16:9" | "9:16" | "4:3" | "3:4" | "3:2" | "2:3" | "4:5" | "5:4" | "21:9" | "9:21" | "97:30" | "adaptive";
-type ImageSize = "1K" | "2K" | "4K";
+type ImageSize = "512" | "1K" | "2K" | "4K";
 type VideoResolution = "360p" | "480p" | "720p" | "768p" | "1080p" | "2k" | "4k";
 /** "xhigh" and "max" exist only on gpt-image-2.5; gpt-image-2 rejects them at the provider. */
 type ImageQuality = "auto" | "low" | "medium" | "high" | "xhigh" | "max";
@@ -37,6 +37,8 @@ interface ImageRequest {
     seed?: number;
     count?: number;
     outputFormat?: OutputFormat;
+    /** Upscale factor for upscaler rows; adapters map it to the provider's field. */
+    scaleFactor?: number;
 }
 /**
  * What a caller passes to generateImage. Same field names as the AI SDK's
@@ -120,7 +122,7 @@ interface MediaPart {
     cleanup?: () => Promise<void>;
 }
 type RequiredInput = "firstFrame" | "sourceVideo" | "references";
-type SupportedOption = "seed" | "quality" | "negativePrompt" | "cfgScale" | "generateAudio" | "promptExpansion" | "lastFrame" | "count" | "outputFormat" | "instrumental";
+type SupportedOption = "seed" | "quality" | "negativePrompt" | "cfgScale" | "generateAudio" | "promptExpansion" | "lastFrame" | "count" | "outputFormat" | "scaleFactor" | "instrumental";
 interface ReferenceLimits {
     images?: number;
     videos?: number;
@@ -208,6 +210,8 @@ interface ImageResult {
     images: ImageBytes[];
     usage: UsageEvent;
     raw?: unknown;
+    /** The provider URL the images were downloaded from, when the provider hands one back. */
+    outputUrl?: string;
 }
 /** What generateImage returns: the adapter result plus `image`, SDK accessors and warnings. */
 interface ImageCallResult extends ImageResult {

package/dist/index.js

@@ -83,7 +83,18 @@ var Catalog = class _Catalog {
 
 // src/catalog/image.ts
 var IMAGE_USAGE = { unit: "images", modality: "IMAGE" };
-var GEMINI_RATIOS = ["1:1", "16:9", "9:16", "4:3", "3:4", "21:9", "4:5", "5:4"];
+var GEMINI_RATIOS = [
+  "1:1",
+  "2:3",
+  "3:2",
+  "16:9",
+  "9:16",
+  "4:3",
+  "3:4",
+  "21:9",
+  "4:5",
+  "5:4"
+];
 function geminiImage(id, label, aliases, sizes) {
   const model = {
     id,
@@ -106,7 +117,8 @@ var IMAGE_MODELS = [
     "gemini-3.1-flash-image-preview",
     "Nano Banana Flash",
     ["nb-2"],
-    ["1K", "2K", "4K"]
+    // Flash 3.1 is the only Gemini image model with the 512 tier.
+    ["512", "1K", "2K", "4K"]
   ),
   geminiImage("gemini-3.1-flash-lite-image", "Nano Banana Lite", ["nb-lite"], ["1K"]),
   // trigger-workflows validates 1K only; pricing seeds 2K/4K rows so allow all three.
@@ -451,8 +463,12 @@ var OTHER_MODELS = [
     adapter: "replicate",
     providerModelId: "topazlabs/image-upscale",
     lifecycle: "active",
-    constraints: {},
-    defaults: {},
+    constraints: {
+      references: { images: 1 },
+      requires: ["references"],
+      supports: ["scaleFactor", "outputFormat"]
+    },
+    defaults: { scaleFactor: 2, outputFormat: "png" },
     usage: { unit: "compute_seconds", modality: "IMAGE" },
     pricingKind: "flat"
   },
@@ -508,7 +524,7 @@ var TEXT_MODELS = [
   text("gpt-5.6-sol", "GPT-5.6 Sol", "gpt-5.6-sol"),
   // Library key "claude-sonnet-4.5" maps to the SDK id "claude-sonnet-4-5".
   text("claude-sonnet-4-5", "Claude Sonnet 4.5", "claude-sonnet-4-5", {
-    aliases: ["claude-sonnet-4.5"]
+    aliases: ["claude-sonnet-4.5", "claude-sonnet-4-5-20250929"]
   }),
   // Anthropic retired the dated 4.0 ids; callers that still send them get 4.6.
   text("claude-sonnet-4-6", "Claude Sonnet 4.6", "claude-sonnet-4-6", {
@@ -524,7 +540,7 @@ var TEXT_MODELS = [
     aliases: ["claude-opus-4.8", "claude-opus-4", "claude-opus-4.0", "claude-opus-4-20250514"]
   }),
   text("claude-haiku-4-5", "Claude Haiku 4.5", "claude-haiku-4-5", {
-    aliases: ["claude-haiku-4.5", "claude-haiku-4-5-20251001"]
+    aliases: ["claude-haiku-4.5", "claude-haiku-4-5-20251001", "claude-haiku-4"]
   }),
   text("claude-opus-5", "Claude Opus 5", "claude-opus-5", {
     aliases: ["claude-opus-5.0"]
@@ -1007,7 +1023,7 @@ var MODELS = [
 ];
 
 // src/resolve-options.ts
-var SIZE_ORDER = ["1K", "2K", "4K"];
+var SIZE_ORDER = ["512", "1K", "2K", "4K"];
 var RESOLUTION_ORDER = ["360p", "480p", "720p", "768p", "1080p", "2k", "4k"];
 function ratioValue(r) {
   if (r === "adaptive") return 1;
@@ -1139,7 +1155,13 @@ function resolveImageOptions(model, request, onUnsupported = "error") {
       model: model.id
     });
   }
-  dropUnsupported(merged, model, warnings, ["seed", "quality", "count", "outputFormat"]);
+  dropUnsupported(merged, model, warnings, [
+    "seed",
+    "quality",
+    "count",
+    "outputFormat",
+    "scaleFactor"
+  ]);
   return { request: merged, warnings };
 }
 function resolveVideoOptions(model, request, onUnsupported = "error") {
@@ -1920,6 +1942,7 @@ function createModelKit(config, adapters) {
       providerOptions,
       references,
       exactSize: exactSizeField,
+      scaleFactor,
       quality: _quality,
       outputFormat: _outputFormat
     } = req;
@@ -1945,6 +1968,8 @@ function createModelKit(config, adapters) {
     if (_quality !== void 0) pushIgnored(warnings, "quality", _quality, model.modelId);
     if (_outputFormat !== void 0)
       pushIgnored(warnings, "outputFormat", _outputFormat, model.modelId);
+    if (scaleFactor !== void 0)
+      pushIgnored(warnings, "scaleFactor", scaleFactor, model.modelId);
     const abortSignal = combineSignals(req.abortSignal, call.signal);
     if (abortSignal) options.abortSignal = abortSignal;
     const result = await aiGenerateImage(options);
@@ -4004,6 +4029,10 @@ function createGoogleImageAdapter(deps = defaultGoogleImageDeps) {
         imageQuality: "auto"
       };
       if (req.size !== void 0) usage2.imageSize = req.size;
+      if (result.usage?.inputTokens !== void 0)
+        usage2.inputTokens = result.usage.inputTokens;
+      if (result.usage?.outputTokens !== void 0)
+        usage2.outputTokens = result.usage.outputTokens;
       return { images, usage: usage2, raw: result };
     }
   };
@@ -4535,7 +4564,7 @@ function aspectRatioToSize(ratio, tier) {
   };
   const MIN_PIXELS = 655360;
   const MAX_PIXELS = 8294400;
-  const maxLong = TIER_LONG_EDGE[tier];
+  const maxLong = TIER_LONG_EDGE[tier] ?? 1024;
   const RATIO_MAP = /* @__PURE__ */ new Map([
     ["1:1", [1, 1]],
     ["16:9", [16, 9]],
@@ -5169,6 +5198,9 @@ function isUpscaler(providerModelId) {
   const id = isLower(providerModelId);
   return id.includes("upscal") || id.includes("crystal") || id.includes("topaz");
 }
+function isTopazUpscaler(providerModelId) {
+  return isLower(providerModelId).includes("topazlabs/image-upscale");
+}
 function isQwen(providerModelId) {
   return isLower(providerModelId).includes("qwen");
 }
@@ -5218,8 +5250,36 @@ function wanImageSize(aspectRatio, size) {
 }
 async function generateReplicateImage(req, ctx, deps) {
   const id = ctx.model.providerModelId;
-  const input = { prompt: req.prompt };
-  if (isUpscaler(id)) {
+  let input;
+  if (isTopazUpscaler(id)) {
+    const first = (req.references ?? [])[0];
+    if (!first) {
+      throw new ModelKitError({
+        code: "missing_input",
+        message: `${ctx.model.id} requires a reference image to upscale`,
+        provider: "replicate",
+        model: ctx.model.id,
+        retryable: false
+      });
+    }
+    let format = req.outputFormat === "jpeg" ? "jpg" : req.outputFormat ?? "png";
+    if (format !== "png" && format !== "jpg") {
+      ctx.warnings.push({
+        code: "option_ignored",
+        field: "outputFormat",
+        from: format,
+        message: "topazlabs/image-upscale supports only png and jpg; using png"
+      });
+      format = "png";
+    }
+    input = {
+      image: refToUrl3(first),
+      enhance_model: "Standard V2",
+      output_format: format,
+      upscale_factor: `${req.scaleFactor ?? 2}x`,
+      face_enhancement: true
+    };
+  } else if (isUpscaler(id)) {
     const refs = req.references ?? [];
     const first = refs[0];
     if (!first) {
@@ -5231,24 +5291,29 @@ async function generateReplicateImage(req, ctx, deps) {
         retryable: false
       });
     }
-    input.image = refToUrl3(first);
-    input.creativity = 0;
-    input.scale_factor = 2;
-    input.output_format = "png";
+    const imageInput = { prompt: req.prompt };
+    imageInput.image = refToUrl3(first);
+    imageInput.creativity = 0;
+    imageInput.scale_factor = 2;
+    imageInput.output_format = "png";
+    input = imageInput;
   } else if (isWanImage(id)) {
     const refs = (req.references ?? []).map(refToUrl3);
-    if (refs.length > 0) input.images = refs;
-    input.size = wanImageSize(req.aspectRatio, req.size);
+    const imageInput = { prompt: req.prompt };
+    if (refs.length > 0) imageInput.images = refs;
+    imageInput.size = wanImageSize(req.aspectRatio, req.size);
+    input = imageInput;
   } else if (isQwen(id)) {
     const refs = (req.references ?? []).map(refToUrl3);
     const firstUrl = refs[0];
+    const imageInput = { prompt: req.prompt };
     if (firstUrl !== void 0) {
-      input.image = firstUrl;
-      if (!isQwenEditPlus(id)) input.match_input_image = true;
+      imageInput.image = firstUrl;
+      if (!isQwenEditPlus(id)) imageInput.match_input_image = true;
     }
     const ratios = isQwenEditPlus(id) ? QWEN_EDIT_RATIOS : QWEN_IMAGE_RATIOS;
     if (req.aspectRatio !== void 0) {
-      if (ratios.has(req.aspectRatio)) input.aspect_ratio = req.aspectRatio;
+      if (ratios.has(req.aspectRatio)) imageInput.aspect_ratio = req.aspectRatio;
       else
         ctx.warnings.push({
           code: "option_ignored",
@@ -5258,18 +5323,21 @@ async function generateReplicateImage(req, ctx, deps) {
         });
     }
     if (isQwenEditPlus(id) && req.outputFormat !== void 0) {
-      input.output_format = req.outputFormat === "jpeg" ? "jpg" : req.outputFormat;
+      imageInput.output_format = req.outputFormat === "jpeg" ? "jpg" : req.outputFormat;
     }
+    input = imageInput;
   } else {
     const refs = req.references ?? [];
     const first = refs[0];
-    if (first) input.image = refToUrl3(first);
+    const imageInput = { prompt: req.prompt };
+    if (first) imageInput.image = refToUrl3(first);
+    input = imageInput;
   }
   const { url, predictTime, raw } = await runPrediction(ctx, input, deps);
   const file = await download(ctx, url);
   const usage2 = imageUsage2(ctx, req);
   usage2.quantity = predictTime;
-  return { images: [{ bytes: file.bytes, mime: file.mime }], usage: usage2, raw };
+  return { images: [{ bytes: file.bytes, mime: file.mime }], usage: usage2, raw, outputUrl: url };
 }
 function wanSize(aspectRatio) {
   if (!aspectRatio || aspectRatio === "adaptive" || aspectRatio === "1:1") return "2K";