@picsart/ai-sdk 6.15.0 → 6.16.0

package/README.md

@@ -153,6 +153,35 @@ const folders = await ai.drive.folders()
 const items = await ai.drive.list()
 ```
 
+Generated Drive assets carry `model` and JSON `aiSDKPayload` automatically. Both
+`generate()` and `submit()` record the original SDK inputs; custom Drive attributes
+cannot replace that provenance. These are requested inputs, not a claim that every
+provider default or rewritten prompt has been recovered. The stored prompt is capped
+at 18,000 characters without changing the prompt sent to generation.
+
+`apis.run()` remains a direct vendor API pass-through. It does not apply the client’s
+Drive configuration or derive SDK metadata, even when a vendor payload explicitly
+requests Drive saving. Vendor-shaped inputs must never be stored as `aiSDKPayload`;
+that attribute is reserved for original SDK-shaped inputs.
+
+For a separate or delayed save, pass the original generation context:
+
+```typescript
+await ai.drive.save({
+  url: generatedUrl,
+  name: 'clip.mp4',
+  resourceType: 'VIDEO',
+  generation: { modelId, params: originalGenerationParams },
+})
+// The existing helper also accepts a full params object instead of just a prompt:
+const saveParams = ai.drive.buildSaveParams(generatedUrl, modelId, modelName, 'video', originalGenerationParams)
+```
+
+Legacy `model`/`prompt`/`textScript` saves are normalized at write time. URL-only
+uploads remain ordinary uploads: without generation context the SDK cannot infer
+model, prompt or settings. Existing Drive files are not backfilled. The backend preserves SDK-supplied metadata; it cannot reconstruct SDK input
+names or catalog model IDs from an arbitrary vendor request.
+
 ## Text Generation (LLMs)
 
 Claude, GPT, and Gemini text models are called with `generateText()`. Single-shot:

package/index.d.ts

@@ -349,14 +349,30 @@ type ModelInputById = {
     "gemini-2.5-flash-tts": {
         language?: string;
         accent?: string;
-        prompt: string;
+        prompt?: string;
         voiceId?: string;
+        parts?: Array<{
+            text: string;
+            speaker?: string;
+        }>;
+        multiSpeakerVoiceConfigs?: Array<{
+            speaker: string;
+            voiceName: string;
+        }>;
     };
     "gemini-2.5-pro-tts": {
         language?: string;
         accent?: string;
-        prompt: string;
+        prompt?: string;
         voiceId?: string;
+        parts?: Array<{
+            text: string;
+            speaker?: string;
+        }>;
+        multiSpeakerVoiceConfigs?: Array<{
+            speaker: string;
+            voiceName: string;
+        }>;
     };
     "gemini-3-pro": {
         prompt: string;
@@ -412,14 +428,34 @@ type ModelInputById = {
     "gemini-3.8-flash-lite-tts": {
         language?: string;
         accent?: string;
-        prompt: string;
+        prompt?: string;
         voiceId?: string;
+        style?: string;
+        parts?: Array<{
+            text: string;
+            speaker?: string;
+            style?: string;
+        }>;
+        multiSpeakerVoiceConfigs?: Array<{
+            speaker: string;
+            voiceName: string;
+        }>;
     };
     "gemini-3.8-flash-tts": {
         language?: string;
         accent?: string;
-        prompt: string;
+        prompt?: string;
         voiceId?: string;
+        style?: string;
+        parts?: Array<{
+            text: string;
+            speaker?: string;
+            style?: string;
+        }>;
+        multiSpeakerVoiceConfigs?: Array<{
+            speaker: string;
+            voiceName: string;
+        }>;
     };
     "gemini-omni-1.1-flash-preview": {
         prompt: string;
@@ -2668,11 +2704,19 @@ interface ListOptions {
     folder?: DriveFolder;
     type?: MediaTypeFilter;
 }
+/** Original generation inputs, portable across clients and delayed Drive saves. */
+interface GenerationProvenance {
+    modelId: string;
+    params: Record<string, unknown>;
+    app?: AppIdentity;
+}
 interface SaveParams {
     url: string;
     name: string;
     resourceType: 'PHOTO' | 'VIDEO' | 'AUDIO';
     attributes?: Record<string, string>;
+    /** Required to retain full provenance when saving a generated URL separately. */
+    generation?: GenerationProvenance;
     previewUrl?: string;
 }
 /** Target folder for backend Drive save. */
@@ -2714,7 +2758,7 @@ interface DriveClient {
     /** Save an asset to Drive. */
     save(params: SaveParams, folder?: DriveFolder): Promise<DriveSaveResult | null>;
     /** Build the save params for a generation result. */
-    buildSaveParams(url: string, modelId: string, modelName: string, mode: string, prompt?: string): SaveParams;
+    buildSaveParams(url: string, modelId: string, modelName: string, mode: string, promptOrParams?: string | Record<string, unknown>): SaveParams;
     /** Set a like/dislike reaction on a file. */
     addReaction(fileUid: string, reaction: UserReaction): Promise<boolean>;
     /** Clear the reaction on a file. */
@@ -2744,12 +2788,7 @@ declare function resolveExtension(mode: string, hints?: FilenameHints): string;
  */
 declare function expectedOutputFormat(model: Pick<ModelDefinition, 'paramConfig' | 'outputExtension'>, params: Record<string, unknown>): string | undefined;
 declare function buildFilename(prompt: string | undefined, mode: string, hints?: FilenameHints): string;
-declare function buildGenerationAttributes(input: {
-    modelId: string;
-    params: Record<string, unknown>;
-    /** TODO(backend-autosave): temporary — remove once the backend stamps appId/appType. */
-    app?: AppIdentity;
-}): DriveAttributes;
+declare function buildGenerationAttributes(input: GenerationProvenance): DriveAttributes;
 declare function parseGeneration(file: DriveFile | Record<string, unknown>): GenerationFile;
 
 /** Result of `ai.apis.run()` — the API result plus optional credit usage. */
@@ -3545,4 +3584,4 @@ declare const getModel: (id: string) => ModelDefinition | undefined;
  */
 declare const findModel: (ref: string) => ModelDefinition | undefined;
 
-export { ALL_MODELS, type AiClient, ApiError, type ApiErrorCode, type ApiErrorInit, type ApiResponse, type ApiRunOptions, type ApiSchemas, type ApisClient, type AppIdentity, type AppType, type AuthenticatedFetch, type AvatarOption, type BooleanDescriptor, type BooleanEntry, type CatalogDescriptor, type CatalogEntry, type CatalogItem, type CatalogPage, type CatalogPageOptions, type CatalogPreview, type CatalogQuery, type CatalogResult, type CatalogSource, type CatalogsClient, type CatalogsOptions, type ClientConfig, type CreditRange, type CreditRangeContext, type CreditTier, type CreditUsage, DEFAULT_VISIBLE_RELEASES, type DeepLinkResult, type DriveAttributes, type DriveClient, type DriveConfig, type DriveDraftInfo, type DriveFile, type DriveFileDetails, type DriveFolder, type DriveMediaItem, type DriveSaveResult, type EntryMeta, type EnumDescriptor, type EnumEntry, type EnumOption, type FileDescriptor, type FileEntry, type FlatParamEntry, type GenerateOptions, type GenerateResult, type GenerateResultItem, type GenerateResultItemMetadata, type GenerateTextResult, type GenerationContext, type GenerationEvent, GenerationEventType, type GenerationFile, type GenerationMode, type GenerationOptions, type GenerationProgress, type ListOptions, type MediaModelId, type MediaTypeFilter, Model, type ModelDefinition, type ModelDescriptor, type ModelFilter, type ModelInput, type ModelInputById, type ModelMeta, type ModelParams, type ModelParamsAccessor, Models, type ObjectDescriptor, type ObjectEntry, type ParamDescriptor, type ParamEntry, type ParamOption, type PayloadDriveFolderOptions, type PayloadDriveOptions, type PayloadInputsTransformationOptions, type PricingOptions, type ProviderInfo, type RangeDescriptor, type RangeEntry, type ReleaseTag, type SaveParams, type SdkPayload, type SdkTransport, type SeedanceDraftTask, type TextDescriptor, type TextEntry, type TextModelId, type TextModelInputById, type ToolUsage, type TransportPollOptions, type TransportResult, type TypedModelId, type UserReaction, type ValidationResult, type VoiceOption, type WorkflowJobHandle, type WorkflowSubmitRequest, buildFilename, buildGenerationAttributes, catalog, createClient, decodeDeepLinkPayload, encodeDeepLinkPayload, expectedOutputFormat, findModel, getModel, getModelsByMode, getVoiceById, inferResourceType, isVisibleForReleases, parseGeneration, releaseOf, resolveExtension, toAvatarOption, toVoiceOption };
+export { ALL_MODELS, type AiClient, ApiError, type ApiErrorCode, type ApiErrorInit, type ApiResponse, type ApiRunOptions, type ApiSchemas, type ApisClient, type AppIdentity, type AppType, type AuthenticatedFetch, type AvatarOption, type BooleanDescriptor, type BooleanEntry, type CatalogDescriptor, type CatalogEntry, type CatalogItem, type CatalogPage, type CatalogPageOptions, type CatalogPreview, type CatalogQuery, type CatalogResult, type CatalogSource, type CatalogsClient, type CatalogsOptions, type ClientConfig, type CreditRange, type CreditRangeContext, type CreditTier, type CreditUsage, DEFAULT_VISIBLE_RELEASES, type DeepLinkResult, type DriveAttributes, type DriveClient, type DriveConfig, type DriveDraftInfo, type DriveFile, type DriveFileDetails, type DriveFolder, type DriveMediaItem, type DriveSaveResult, type EntryMeta, type EnumDescriptor, type EnumEntry, type EnumOption, type FileDescriptor, type FileEntry, type FlatParamEntry, type GenerateOptions, type GenerateResult, type GenerateResultItem, type GenerateResultItemMetadata, type GenerateTextResult, type GenerationContext, type GenerationEvent, GenerationEventType, type GenerationFile, type GenerationMode, type GenerationOptions, type GenerationProgress, type GenerationProvenance, type ListOptions, type MediaModelId, type MediaTypeFilter, Model, type ModelDefinition, type ModelDescriptor, type ModelFilter, type ModelInput, type ModelInputById, type ModelMeta, type ModelParams, type ModelParamsAccessor, Models, type ObjectDescriptor, type ObjectEntry, type ParamDescriptor, type ParamEntry, type ParamOption, type PayloadDriveFolderOptions, type PayloadDriveOptions, type PayloadInputsTransformationOptions, type PricingOptions, type ProviderInfo, type RangeDescriptor, type RangeEntry, type ReleaseTag, type SaveParams, type SdkPayload, type SdkTransport, type SeedanceDraftTask, type TextDescriptor, type TextEntry, type TextModelId, type TextModelInputById, type ToolUsage, type TransportPollOptions, type TransportResult, type TypedModelId, type UserReaction, type ValidationResult, type VoiceOption, type WorkflowJobHandle, type WorkflowSubmitRequest, buildFilename, buildGenerationAttributes, catalog, createClient, decodeDeepLinkPayload, encodeDeepLinkPayload, expectedOutputFormat, findModel, getModel, getModelsByMode, getVoiceById, inferResourceType, isVisibleForReleases, parseGeneration, releaseOf, resolveExtension, toAvatarOption, toVoiceOption };

package/index.js

@@ -6453,11 +6453,6 @@ var buildGemini31FlashLiteImagePayload = (ctx) => ({
   imageSize: ctx.resolution ?? "1K",
   ...buildThinkingConfig(ctx)
 });
-var buildGeminiTTSPayload = (model) => (ctx) => ({
-  text: ctx.prompt,
-  model,
-  voiceName: ctx.voiceId ?? "Kore"
-});
 function inferMimeType2(url) {
   return url.match(/\.png(\?|$)/i) ? "image/png" : "image/jpeg";
 }
@@ -6469,6 +6464,52 @@ var buildGeminiOmniVideoPayload = (ctx) => ({
   ...ctx.imageUrls?.[0] ? { image: { url: ctx.imageUrls[0], mimeType: inferMimeType2(ctx.imageUrls[0]) } } : {},
   ...ctx.videoUrl ? { video: { url: ctx.videoUrl } } : {}
 });
+var GEMINI_TTS_TEXT_MAX = 6e3;
+var GEMINI_TTS_STYLE_MAX = 500;
+var geminiTtsParams = (directable) => ({
+  ...params.language(true),
+  // 6,000 boundary-verified against the live gemini/v1/audios worker
+  // (accepts >5,000; see scripts/api-tests/audio-charlimit-boundary-probe.mjs).
+  ...params.prompt({ required: false, maxLength: GEMINI_TTS_TEXT_MAX }),
+  ...params.voiceId([], GEMINI_DEFAULT_VOICE_ID, { catalog: { workflow: "gemini/v1/catalog/voices" } }),
+  ...directable ? {
+    style: {
+      label: "Style",
+      descriptor: {
+        kind: "text",
+        maxLength: GEMINI_TTS_STYLE_MAX,
+        placeholder: "e.g. warm, unhurried; a bedtime story"
+      }
+    }
+  } : {},
+  // Worker `SpeechPartDto[]`, same names. Used instead of `prompt`.
+  parts: {
+    label: "Spoken Parts",
+    descriptor: {
+      kind: "object",
+      array: { min: 1, max: 200 },
+      fields: {
+        text: { label: "Text", kind: "text", maxLength: GEMINI_TTS_TEXT_MAX },
+        speaker: { label: "Speaker", kind: "text", required: false },
+        ...directable ? { style: { label: "Style", kind: "text", maxLength: GEMINI_TTS_STYLE_MAX, required: false } } : {}
+      }
+    }
+  },
+  // Worker `SpeakerVoiceConfigDto[]`, max 2. Replaces `voiceId`. The voice is
+  // a plain id, not a catalog picker: catalog loading resolves top-level params
+  // only, so a caller lists voices with `ai.catalogs.voices(<model>)`.
+  multiSpeakerVoiceConfigs: {
+    label: "Speakers",
+    descriptor: {
+      kind: "object",
+      array: { min: 1, max: 2 },
+      fields: {
+        speaker: { label: "Speaker", kind: "text", placeholder: "Speaker 1" },
+        voiceName: { label: "Voice", kind: "text", placeholder: GEMINI_DEFAULT_VOICE_ID }
+      }
+    }
+  }
+});
 var GEMINI_AR_WIDE = ["1:1", "16:9", "9:16", "3:4", "4:3", "2:3", "21:9", "auto"];
 var thinkingLevelParam = {
   thinkingLevel: {
@@ -6586,32 +6627,35 @@ var { MODELS: MODELS22 } = defineModels("google", [
     }
   },
   // ── Audio ─────────────────────────────────────────────────────────
+  // Payload builders live in gemini.payloads.ts (typed against ModelInput).
+  // The wire `model` is fixed per entry there: `modelId` is the pricing key and
+  // never reaches the builder.
+  //
+  // 2.5 and 3.8 take direction and speakers differently, so the builders
+  // resolve them per generation. 3.8 reads `style` and per-part speakers from
+  // `speechMetadata` and speaks a natural-language prefix in the text aloud:
+  // `parts` go through and `language` / `accent` fold into `style`. 2.5
+  // ignores `speechMetadata`: it reads speakers from "Speaker: line" labels
+  // and direction from a "Say …:" prefix, so the builder writes both into
+  // one text.
   {
     id: "gemini-2.5-flash-tts",
     name: "Gemini 2.5 Flash TTS",
     addedAt: "2026-02-15",
     workflow: "gemini/v1/audios",
-    buildPayload: buildGeminiTTSPayload("gemini-2.5-flash-tts"),
     estimatedTime: 15,
     mode: "audio",
     inputType: "tts",
     modelId: "gemini-2.5-flash-tts",
     description: "Google Gemini native text-to-speech with expressive multilingual voices.",
-    features: [feat("Multilingual", "characteristic"), feat("30 Voices", "characteristic")],
-    paramConfig: {
-      ...params.language(true),
-      // 6,000 boundary-verified against the live gemini/v1/audios worker
-      // (accepts >5,000; see scripts/api-tests/audio-charlimit-boundary-probe.mjs).
-      ...params.prompt({ maxLength: 6e3 }),
-      ...params.voiceId([], GEMINI_DEFAULT_VOICE_ID, { catalog: { workflow: "gemini/v1/catalog/voices" } })
-    }
+    features: [feat("Multilingual", "characteristic"), feat("30 Voices", "characteristic"), feat("Multi-Speaker", "characteristic")],
+    paramConfig: geminiTtsParams(false)
   },
   {
     id: "gemini-2.5-pro-tts",
     name: "Gemini 2.5 Pro TTS",
     addedAt: "2026-03-18",
     workflow: "gemini/v1/audios",
-    buildPayload: buildGeminiTTSPayload("gemini-2.5-pro-tts"),
     estimatedTime: 20,
     mode: "audio",
     inputType: "tts",
@@ -6619,54 +6663,34 @@ var { MODELS: MODELS22 } = defineModels("google", [
     badge: ["premium"],
     description: "Premium Gemini TTS with richer expressiveness and multi-speaker support.",
     features: [feat("Multilingual", "characteristic"), feat("30 Voices", "characteristic"), feat("Multi-Speaker", "characteristic")],
-    paramConfig: {
-      ...params.language(true),
-      // 6,000 boundary-verified against the live gemini/v1/audios worker
-      // (accepts >5,000; see scripts/api-tests/audio-charlimit-boundary-probe.mjs).
-      ...params.prompt({ maxLength: 6e3 }),
-      ...params.voiceId([], GEMINI_DEFAULT_VOICE_ID, { catalog: { workflow: "gemini/v1/catalog/voices" } })
-    }
+    paramConfig: geminiTtsParams(false)
   },
   {
     id: "gemini-3.8-flash-tts",
     name: "Gemini 3.8 Flash TTS",
     addedAt: "2026-09-24",
     workflow: "gemini/v1/audios",
-    buildPayload: buildGeminiTTSPayload("gemini-3.8-flash-tts"),
     estimatedTime: 15,
     mode: "audio",
     inputType: "tts",
     modelId: "gemini-3.8-flash-tts",
-    description: "Latest Gemini native text-to-speech with expressive multilingual voices.",
-    features: [feat("Multilingual", "characteristic"), feat("30 Voices", "characteristic")],
-    paramConfig: {
-      ...params.language(true),
-      // 6,000 boundary-verified against the live gemini/v1/audios worker
-      // (accepts >5,000; see scripts/api-tests/audio-charlimit-boundary-probe.mjs).
-      ...params.prompt({ maxLength: 6e3 }),
-      ...params.voiceId([], GEMINI_DEFAULT_VOICE_ID, { catalog: { workflow: "gemini/v1/catalog/voices" } })
-    }
+    description: "Latest Gemini native text-to-speech with directable delivery and inline vocal events like <laugh> and <sigh>.",
+    features: [feat("Multilingual", "characteristic"), feat("30 Voices", "characteristic"), feat("Multi-Speaker", "characteristic"), feat("Style Direction", "characteristic")],
+    paramConfig: geminiTtsParams(true)
   },
   {
     id: "gemini-3.8-flash-lite-tts",
     name: "Gemini 3.8 Flash Lite TTS",
     addedAt: "2026-09-24",
     workflow: "gemini/v1/audios",
-    buildPayload: buildGeminiTTSPayload("gemini-3.8-flash-lite-tts"),
     estimatedTime: 12,
     mode: "audio",
     inputType: "tts",
     modelId: "gemini-3.8-flash-lite-tts",
     badge: ["fast"],
-    description: "Lightweight Gemini 3.8 TTS variant for faster, high-volume speech synthesis.",
-    features: [feat("Multilingual", "characteristic"), feat("30 Voices", "characteristic")],
-    paramConfig: {
-      ...params.language(true),
-      // 6,000 boundary-verified against the live gemini/v1/audios worker
-      // (accepts >5,000; see scripts/api-tests/audio-charlimit-boundary-probe.mjs).
-      ...params.prompt({ maxLength: 6e3 }),
-      ...params.voiceId([], GEMINI_DEFAULT_VOICE_ID, { catalog: { workflow: "gemini/v1/catalog/voices" } })
-    }
+    description: "Lightweight Gemini 3.8 TTS with directable delivery for faster, high-volume speech synthesis.",
+    features: [feat("Multilingual", "characteristic"), feat("30 Voices", "characteristic"), feat("Multi-Speaker", "characteristic"), feat("Style Direction", "characteristic")],
+    paramConfig: geminiTtsParams(true)
   },
   // ── Video ─────────────────────────────────────────────────────────
   {
@@ -6733,8 +6757,95 @@ var buildOmniFlash11Payload = (input) => ({
   ...input.videoUrl ? { video: { url: input.videoUrl } } : {},
   ...input.videoUrls?.length ? { referenceVideos: input.videoUrls.map((url) => ({ url })) } : {}
 });
+var STYLE_MAX = 500;
+var TEXT_MAX = 6e3;
+var invalid = (message) => new ApiError(message, { status: 400, code: "validation_error" });
+var blank = (value) => !value?.trim();
+function checkInput(input) {
+  const configs = input.multiSpeakerVoiceConfigs ?? [];
+  for (const [i, config] of configs.entries()) {
+    if (blank(config.speaker)) throw invalid(`Gemini TTS: speaker config ${i + 1} has no speaker.`);
+    if (blank(config.voiceName)) throw invalid(`Gemini TTS: speaker config ${i + 1} has no voiceName.`);
+  }
+  const speakers = configs.map((config) => config.speaker);
+  if (new Set(speakers).size !== speakers.length) {
+    throw invalid("Gemini TTS: multiSpeakerVoiceConfigs names the same speaker twice.");
+  }
+  const parts = input.parts?.length ? input.parts : void 0;
+  if (!parts) {
+    if (blank(input.prompt)) throw invalid("Gemini TTS: provide a prompt or parts.");
+    return { speakers };
+  }
+  for (const [i, part] of parts.entries()) {
+    if (blank(part.text)) throw invalid(`Gemini TTS: part ${i + 1} has no text.`);
+    if (speakers.length && (!part.speaker || !speakers.includes(part.speaker))) {
+      throw invalid(
+        `Gemini TTS: part ${i + 1} needs a speaker from multiSpeakerVoiceConfigs (${speakers.join(", ")}).`
+      );
+    }
+    if (!speakers.length && part.speaker) {
+      throw invalid(`Gemini TTS: part ${i + 1} names a speaker but no multiSpeakerVoiceConfigs are set.`);
+    }
+  }
+  return { parts, speakers };
+}
+function assertLength(spoken) {
+  if (spoken.length > TEXT_MAX) {
+    throw invalid(`Gemini TTS: the text totals ${spoken.length} characters; the limit is ${TEXT_MAX}.`);
+  }
+}
+var voiceFields = (input) => input.multiSpeakerVoiceConfigs?.length ? { multiSpeakerVoiceConfigs: input.multiSpeakerVoiceConfigs } : { voiceName: input.voiceId ?? GEMINI_DEFAULT_VOICE_ID };
+function composeStyle(input) {
+  const language = input.language?.trim();
+  const accent = input.accent?.trim();
+  const voice = language || accent ? `speak${language ? ` in ${language}` : ""}${accent ? ` with a ${accent} accent` : ""}` : void 0;
+  const composed = [...input.styles.map((style) => style?.trim()), voice].filter(Boolean).join("; ") || void 0;
+  if (composed && composed.length > STYLE_MAX) {
+    throw invalid(`Gemini TTS: style with language/accent exceeds ${STYLE_MAX} characters.`);
+  }
+  return composed;
+}
+function directionPrefix(input) {
+  const language = input.language?.trim();
+  const accent = input.accent?.trim();
+  if (!language && !accent) return "";
+  return `Say${language ? ` in ${language}` : ""}${accent ? ` with a ${accent} accent` : ""}: `;
+}
+var buildTts25Payload = (model) => (input) => {
+  const { parts, speakers } = checkInput(input);
+  const body = parts ? parts.map((part) => speakers.length ? `${part.speaker}: ${part.text}` : part.text).join("\n") : input.prompt;
+  const text = `${directionPrefix(input)}${body}`;
+  assertLength(text);
+  return { text, model, ...voiceFields(input) };
+};
+var buildTts38Payload = (model) => (input) => {
+  const { parts } = checkInput(input);
+  const direction = { language: input.language, accent: input.accent };
+  if (!parts) {
+    assertLength(input.prompt);
+    const style = composeStyle({ styles: [input.style], ...direction });
+    return { text: input.prompt, model, ...voiceFields(input), ...style ? { style } : {} };
+  }
+  assertLength(parts.map((part) => part.text).join("\n"));
+  return {
+    parts: parts.map((part) => {
+      const style = composeStyle({ styles: [input.style, part.style], ...direction });
+      return {
+        text: part.text,
+        ...part.speaker ? { speaker: part.speaker } : {},
+        ...style ? { style } : {}
+      };
+    }),
+    model,
+    ...voiceFields(input)
+  };
+};
 registerPayloads(MODELS22, {
-  "gemini-omni-1.1-flash-preview": buildOmniFlash11Payload
+  "gemini-omni-1.1-flash-preview": buildOmniFlash11Payload,
+  "gemini-2.5-flash-tts": buildTts25Payload("gemini-2.5-flash-tts"),
+  "gemini-2.5-pro-tts": buildTts25Payload("gemini-2.5-pro-tts"),
+  "gemini-3.8-flash-tts": buildTts38Payload("gemini-3.8-flash-tts"),
+  "gemini-3.8-flash-lite-tts": buildTts38Payload("gemini-3.8-flash-lite-tts")
 });
 
 // src/vendors/catalog/openai.ts
@@ -11997,6 +12108,23 @@ function buildGenerationAttributes(input) {
   }
   return attrs;
 }
+function attributesForSave(params2) {
+  const attributes = { ...params2.attributes };
+  if (params2.generation) {
+    return { ...attributes, ...buildGenerationAttributes(params2.generation) };
+  }
+  if (asString(attributes.model)) {
+    const legacy = adaptLegacyGeneration(attributes);
+    const canonical = parseJsonAttr(attributes.aiSDKPayload);
+    const payload = canonical && typeof canonical.prompt === "string" ? canonical : legacy.aiSDKPayload ?? {};
+    return {
+      ...legacy.appId ? { appId: legacy.appId, appType: legacy.appType } : {},
+      ...attributes,
+      ...buildGenerationAttributes({ modelId: attributes.model, params: payload })
+    };
+  }
+  return attributes;
+}
 function toMediaItem(file) {
   const url = normalizeUrl(file.sourceUrl);
   if (!url || String(file.name || "").startsWith("__")) return null;
@@ -12332,7 +12460,7 @@ function createDriveClient(f, apiUrl, rootFolderName) {
           width: 1024,
           height: 1024
         },
-        attributes: Object.entries(params2.attributes ?? {}).map(([property, value]) => ({
+        attributes: Object.entries(attributesForSave(params2)).map(([property, value]) => ({
           property,
           value
         }))
@@ -12360,15 +12488,17 @@ function createDriveClient(f, apiUrl, rootFolderName) {
       }
     },
     /** Build standard save params from a generation result. */
-    buildSaveParams(url, modelId, modelName, mode, prompt) {
+    buildSaveParams(url, modelId, modelName, mode, promptOrParams) {
+      const params2 = typeof promptOrParams === "string" ? { prompt: promptOrParams } : promptOrParams ?? {};
+      const prompt = String(params2.prompt ?? "");
       return {
         url,
         name: buildFilename(prompt, mode, { url }),
         resourceType: inferResourceType(mode),
         attributes: {
           tool: "ai-sdk",
-          model: modelId,
-          prompt: prompt || "",
+          ...buildGenerationAttributes({ modelId, params: params2 }),
+          prompt,
           service: modelName
         }
       };
@@ -12676,8 +12806,8 @@ function createClient(config) {
     return {
       // Named before the job runs, so the requested/declared format is the best hint.
       name: explicit?.name ?? buildFilename(params2.prompt, model.mode, { format: expectedOutputFormat(model, params2) }),
-      // SDK-assembled attributes are the baseline; explicit attributes win per-key.
-      attributes: { ...attributes, ...explicit?.attributes ?? {} },
+      // Generation provenance is authoritative; custom attributes cannot replace it.
+      attributes: { ...explicit?.attributes ?? {}, ...attributes },
       folder: explicit?.folder ?? (folderPath ? { path: folderPath } : void 0)
     };
   }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@picsart/ai-sdk",
-  "version": "6.15.0",
+  "version": "6.16.0",
   "type": "module",
   "description": "Type-safe SDK for 100+ AI models — image, video, audio, and text generation with Picsart",
   "license": "MIT",