mulmocast 2.15.0 → 2.17.0

package/README.md

@@ -103,11 +103,6 @@ Create a `.env` file in your project directory with the following API keys:
 OPENAI_API_KEY=your_openai_api_key
 ```
 
-#### (Optional) For the advanced image generation model
-```bash
-DEFAULT_OPENAI_IMAGE_MODEL=gpt-image-1 # for the advanced image generation model
-```
-
 #### (Optional) For Google's image and TTS.
 ```bash
 GEMINI_API_KEY=your_google_gemini_api_key
@@ -298,10 +293,9 @@ Optionally, you can specify *__clipboard* as the script file name to paste the s
 Verify your `.env` file contains:
 ```bash
 OPENAI_API_KEY=your_openai_api_key
-DEFAULT_OPENAI_IMAGE_MODEL=gpt-image-1 # required for high-quality Ghibli-style images
 ```
 
-> **Note:** Ensure your OpenAI organization is verified to access the gpt-image-1 model. Visit https://platform.openai.com/settings/organization/general and complete the "Verifications" section.
+> **Note:** Ensure your OpenAI organization is verified to access the GPT Image models. Visit https://platform.openai.com/settings/organization/general and complete the "Verifications" section.
 
 ### Step 2: Generate a Ghibli-style MulmoScript
 ```bash
@@ -410,7 +404,7 @@ Payload shape:
   "records": 3,
   "byModel": [
     { "provider": "openai",     "model": "gpt-4o-mini-tts", "records": 1, "inputChars": 20 },
-    { "provider": "gemini",     "model": "gemini-2.5-flash-preview-tts", "records": 1, "inputTokens": 9, "outputTokens": 59, "totalTokens": 68 },
+    { "provider": "gemini",     "model": "gemini-3.8-flash-tts", "records": 1, "inputTokens": 9, "outputTokens": 59, "totalTokens": 68 },
     { "provider": "elevenlabs", "model": "eleven_multilingual_v2", "records": 1, "inputChars": 18 }
   ],
   "snapshot": [ /* per-call UsageRecord[] */ ]

package/lib/actions/image_references.d.ts

@@ -1,6 +1,30 @@
-import { MulmoStudioContext, MulmoBeat, MulmoImagePromptMedia } from "../types/index.js";
+import { MulmoPresentationStyleMethods } from "../methods/index.js";
+import { MulmoStudioContext, MulmoBeat, MulmoImagePromptMedia, MulmoCanvasDimension } from "../types/index.js";
 import type { ResolvedImageReference } from "../methods/mulmo_image_prompt_media.js";
 export type { ResolvedImageReference } from "../methods/mulmo_image_prompt_media.js";
+type ImageAgentInfo = ReturnType<typeof MulmoPresentationStyleMethods.getImageAgentInfo>;
+type MovieAgentInfo = ReturnType<typeof MulmoPresentationStyleMethods.getMovieAgentInfo>;
+export declare const referenceImageGraphParams: (imageAgentInfo: ImageAgentInfo, canvasSize: MulmoCanvasDimension) => {
+    model: string | undefined;
+    moderation: string | undefined;
+    canvasSize: {
+        width: number;
+        height: number;
+    };
+    quality: string | undefined;
+    vertexai_project: string | undefined;
+    vertexai_location: string | undefined;
+};
+export declare const referenceMovieGraphParams: (movieAgentInfo: MovieAgentInfo, canvasSize: MulmoCanvasDimension) => {
+    model: string | undefined;
+    canvasSize: {
+        width: number;
+        height: number;
+    };
+    generateAudio: boolean | undefined;
+    vertexai_project: string | undefined;
+    vertexai_location: string | undefined;
+};
 export declare const generateReferenceImage: (inputs: {
     context: MulmoStudioContext;
     key: string;

package/lib/actions/image_references.js

@@ -6,6 +6,22 @@ import { MulmoPresentationStyleMethods, MulmoMediaSourceMethods, MulmoImagePromp
 import { ImageMediaType, } from "../types/index.js";
 import { imageOpenaiAgent, mediaMockAgent, imageGenAIAgent, imageReplicateAgent, movieGenAIAgent, movieReplicateAgent } from "../agents/index.js";
 import { agentGenerationError, imageReferenceAction, imageFileTarget, movieFileTarget } from "../utils/error_cause.js";
+// Reference media are generated outside the beat graphs, so they must carry the same params the beats do.
+export const referenceImageGraphParams = (imageAgentInfo, canvasSize) => ({
+    model: imageAgentInfo.imageParams.model,
+    moderation: imageAgentInfo.imageParams.moderation,
+    canvasSize,
+    quality: imageAgentInfo.imageParams.quality,
+    vertexai_project: imageAgentInfo.imageParams.vertexai_project,
+    vertexai_location: imageAgentInfo.imageParams.vertexai_location,
+});
+export const referenceMovieGraphParams = (movieAgentInfo, canvasSize) => ({
+    model: movieAgentInfo.movieParams.model,
+    canvasSize,
+    generateAudio: movieAgentInfo.movieParams.generateAudio,
+    vertexai_project: movieAgentInfo.movieParams.vertexai_project,
+    vertexai_location: movieAgentInfo.movieParams.vertexai_location,
+});
 // Resolve the normalized reference list to local file paths (with labels), in reference order.
 // imageKey is the imageRefs key of the image being generated (used for warnings and url
 // download file naming). A named reference that is missing from imageRefs warns and is
@@ -62,10 +78,7 @@ export const generateReferenceImage = async (inputs) => {
                         sessionType: "imageReference",
                     },
                 },
-                params: {
-                    model: imageAgentInfo.imageParams.model,
-                    canvasSize: image.canvasSize ?? context.presentationStyle.canvasSize,
-                },
+                params: referenceImageGraphParams(imageAgentInfo, image.canvasSize ?? context.presentationStyle.canvasSize),
             },
         },
     };
@@ -145,11 +158,7 @@ const generateReferenceMovie = async (inputs) => {
                         sessionType: "imageReference",
                     },
                 },
-                params: {
-                    model: movieAgentInfo.movieParams.model,
-                    canvasSize: context.presentationStyle.canvasSize,
-                    generateAudio: movieAgentInfo.movieParams.generateAudio,
-                },
+                params: referenceMovieGraphParams(movieAgentInfo, context.presentationStyle.canvasSize),
             },
         },
     };

package/lib/agents/image_genai_agent.js

@@ -1,6 +1,6 @@
 import fs from "fs";
 import { GraphAILogger } from "graphai";
-import { provider2ImageAgent, deprecatedGoogleImageModelHints, vertexAIGlobalOnlyImageModels, } from "../types/provider2agent.js";
+import { provider2ImageAgent, deprecatedGoogleImageModelHints, vertexAIGlobalOnlyImageModels, isGeminiImageModel, vertexImageLocation, } from "../types/provider2agent.js";
 import { apiKeyMissingError, agentIncorrectAPIKeyError, agentGenerationError, agentInvalidResponseError, imageAction, imageFileTarget, unsupportedModelTarget, hasCause, getGenAIErrorReason, resultify, } from "../utils/error_cause.js";
 import { getAspectRatio } from "../utils/utils.js";
 import { ASPECT_RATIOS, PRO_ASPECT_RATIOS } from "../types/const.js";
@@ -85,7 +85,7 @@ export const imageGenAIAgent = async ({ namedInputs, params, config, }) => {
     const apiKey = config?.apiKey;
     const ai = params.vertexai_project
         ? (() => {
-            const location = params.vertexai_location ?? "us-central1";
+            const location = vertexImageLocation(model, params.vertexai_location);
             if (vertexAIGlobalOnlyImageModels.has(model) && location !== "global") {
                 GraphAILogger.warn(`imageGenAIAgent: model "${model}" on Vertex AI is only available in location "global", but got "${location}". Set imageParams.vertexai_location to "global".`);
             }
@@ -99,7 +99,7 @@ export const imageGenAIAgent = async ({ namedInputs, params, config, }) => {
             }
             return new GoogleGenAI({ apiKey, httpOptions: { timeout: GENAI_REQUEST_TIMEOUT_MS } });
         })();
-    if (model === "gemini-2.5-flash-image" || model === "gemini-3.1-flash-image-preview" || model === "gemini-3-pro-image-preview") {
+    if (isGeminiImageModel(model)) {
         const contentParams = (() => {
             const contents = getGeminiContents(prompt, referenceImages);
             return {

package/lib/agents/image_openai_agent.js

@@ -4,7 +4,7 @@ import { GraphAILogger } from "graphai";
 import { toFile, AuthenticationError, RateLimitError, APIError } from "openai";
 import { createOpenAIClient } from "../utils/openai_client.js";
 import { safeFetch, FETCH_DOWNLOAD_TIMEOUT_MS } from "../utils/fetch.js";
-import { provider2ImageAgent, gptImages, deprecatedOpenAIImageModelHints } from "../types/provider2agent.js";
+import { provider2ImageAgent, gptImages, gptImageQuality, deprecatedOpenAIImageModelHints } from "../types/provider2agent.js";
 import { apiKeyMissingError, agentGenerationError, openAIAgentGenerationError, agentIncorrectAPIKeyError, agentAPIRateLimitError, agentInvalidResponseError, imageAction, imageFileTarget, unsupportedModelTarget, } from "../utils/error_cause.js";
 const isDeprecatedOpenAIImageModel = (model) => model in deprecatedOpenAIImageModelHints;
 export const buildDeprecatedModelMessage = (model) => {
@@ -63,8 +63,9 @@ export const imageOpenaiAgent = async ({ namedInputs, params, config, }) => {
     if (gptImages.includes(model)) {
         imageOptions.moderation = moderation || "auto";
         imageOptions.background = "opaque";
-        if (quality) {
-            imageOptions.quality = quality;
+        const resolvedQuality = gptImageQuality(model, quality);
+        if (resolvedQuality) {
+            imageOptions.quality = resolvedQuality;
         }
     }
     const response = await (async () => {

package/lib/agents/movie_genai_agent.js

@@ -5,7 +5,7 @@ import { apiKeyMissingError, agentGenerationError, agentInvalidResponseError, im
 import { getAspectRatio } from "../utils/utils.js";
 import { ffmpegGetMediaDuration } from "../utils/ffmpeg_utils.js";
 import { ASPECT_RATIOS } from "../types/const.js";
-import { getModelDuration, provider2MovieAgent, AUDIO_MODE_NEVER, AUDIO_MODE_ALWAYS } from "../types/provider2agent.js";
+import { getModelDuration, provider2MovieAgent, AUDIO_MODE_NEVER, AUDIO_MODE_ALWAYS, defaultMovieModel, veoExtensionCount, VEO_EXTENSION_MODEL, VEO_SEGMENT_SEC, } from "../types/provider2agent.js";
 // Per-request timeout so a stalled GenAI video API call rejects instead of hanging.
 const GENAI_REQUEST_TIMEOUT_MS = 120_000;
 // Wall-clock cap on the long-running video operation poll loop (Veo runs minutes).
@@ -89,14 +89,12 @@ const createVeo31Payload = (model, prompt, aspectRatio, source) => ({
     ...source,
 });
 const generateExtendedVideo = async (ai, model, prompt, aspectRatio, imagePath, requestedDuration, movieFile, isVertexAI) => {
-    const initialDuration = 8;
-    const maxExtensionDuration = 8;
-    const extensionsNeeded = Math.ceil((requestedDuration - initialDuration) / maxExtensionDuration);
+    const extensionsNeeded = veoExtensionCount(requestedDuration);
     GraphAILogger.info(`Veo 3.1 video extension: ${extensionsNeeded} extensions needed for ${requestedDuration}s target`);
     const generateIteration = async (iteration, accumulatedDuration, previousVideo) => {
         const isInitial = iteration === 0;
         const remainingDuration = requestedDuration - accumulatedDuration;
-        const extensionDuration = isInitial ? initialDuration : (getModelDuration("google", model, remainingDuration) ?? maxExtensionDuration);
+        const extensionDuration = isInitial ? VEO_SEGMENT_SEC : (getModelDuration("google", model, remainingDuration) ?? VEO_SEGMENT_SEC);
         const getSource = () => {
             if (isInitial)
                 return imagePath ? { image: loadImageAsBase64(imagePath) } : undefined;
@@ -172,7 +170,7 @@ const generateStandardVideo = async (ai, model, prompt, aspectRatio, imagePath,
 export const movieGenAIAgent = async ({ namedInputs, params, config, }) => {
     const { prompt, imagePath, lastFrameImagePath, referenceImages, movieFile } = namedInputs;
     const aspectRatio = getAspectRatio(params.canvasSize, ASPECT_RATIOS);
-    const model = params.model ?? provider2MovieAgent.google.defaultModel;
+    const model = params.model ?? defaultMovieModel("google", params);
     const apiKey = config?.apiKey;
     try {
         const requestedDuration = params.duration ?? 8;
@@ -211,7 +209,7 @@ export const movieGenAIAgent = async ({ namedInputs, params, config, }) => {
                 return new GoogleGenAI({ apiKey, httpOptions: { timeout: GENAI_REQUEST_TIMEOUT_MS } });
             })();
         // Veo 3.1: Video extension mode for videos longer than 8s
-        if (model === "veo-3.1-generate-preview" && requestedDuration > 8 && params.canvasSize) {
+        if (model === VEO_EXTENSION_MODEL && requestedDuration > VEO_SEGMENT_SEC && params.canvasSize) {
             return generateExtendedVideo(ai, model, prompt, aspectRatio, imagePath, requestedDuration, movieFile, isVertexAI);
         }
         // Standard mode

package/lib/agents/tts_gemini_agent.d.ts

@@ -1,5 +1,61 @@
 import type { AgentFunction, AgentFunctionInfo } from "graphai";
+import { type GenerateContentConfig } from "@google/genai";
+import { buildGeminiInteractionsTtsRequest, type GeminiInteractionsTtsInput } from "../utils/gemini_tts.js";
 import type { GoogleTTSAgentParams, AgentBufferResult, AgentTextInputs, AgentErrorResult } from "../types/agent.js";
+import type { AgentUsage } from "../types/usage.js";
+type GeminiAudio = {
+    rawPcm: Buffer;
+    sampleRate: number;
+    usage: AgentUsage | undefined;
+};
+type InlineAudioResponse = {
+    candidates?: {
+        content?: {
+            parts?: {
+                inlineData?: {
+                    data?: string;
+                    mimeType?: string;
+                };
+            }[];
+        };
+    }[];
+    usageMetadata?: {
+        promptTokenCount?: number;
+        candidatesTokenCount?: number;
+        totalTokenCount?: number;
+    };
+};
+type InteractionAudioResponse = {
+    output_audio?: {
+        data?: string;
+        mime_type?: string;
+        sample_rate?: number;
+    };
+    usage?: {
+        total_input_tokens?: number;
+        total_output_tokens?: number;
+        total_tokens?: number;
+    };
+};
+export type GeminiTtsClient = {
+    models: {
+        generateContent: (params: {
+            model: string;
+            contents: {
+                parts: {
+                    text: string;
+                }[];
+            }[];
+            config: GenerateContentConfig;
+        }) => Promise<InlineAudioResponse>;
+    };
+    interactions: {
+        create: (params: ReturnType<typeof buildGeminiInteractionsTtsRequest>, options?: {
+            timeout?: number;
+        }) => Promise<InteractionAudioResponse>;
+    };
+};
+export declare const generateGeminiTts: (ai: GeminiTtsClient, input: GeminiInteractionsTtsInput) => Promise<GeminiAudio>;
 export declare const ttsGeminiAgent: AgentFunction<GoogleTTSAgentParams, AgentBufferResult | AgentErrorResult, AgentTextInputs>;
 declare const ttsGeminiAgentInfo: AgentFunctionInfo;
 export default ttsGeminiAgentInfo;

package/lib/agents/tts_gemini_agent.js

@@ -3,16 +3,51 @@ import { GoogleGenAI } from "@google/genai";
 import { provider2TTSAgent } from "../types/provider2agent.js";
 import { agentIncorrectAPIKeyError, apiKeyMissingError, agentGenerationError, audioAction, audioFileTarget, getGenAIErrorReason, } from "../utils/error_cause.js";
 import { pcmToMp3 } from "../utils/ffmpeg_utils.js";
+import { GEMINI_TTS_SAMPLE_RATE, buildGeminiInteractionsTtsRequest, geminiDirectorsNotesPrompt, pcmFromInteractionAudio, usesGeminiInteractionsTts, } from "../utils/gemini_tts.js";
 // Per-request timeout so a stalled GenAI TTS call rejects (and GraphAI retry can
 // recover) instead of hanging. Generous vs. normal generation time.
 const GENAI_REQUEST_TIMEOUT_MS = 120_000;
-const getPrompt = (text, instructions) => {
-    // https://ai.google.dev/gemini-api/docs/speech-generation?hl=ja#controllable
-    if (instructions) {
-        return `### DIRECTOR'S NOTES\n${instructions}\n\n#### TRANSCRIPT\n${text}`;
-    }
-    return text;
+const geminiUsage = (model, inputTokens, outputTokens, totalTokens) => ({
+    provider: "gemini",
+    model,
+    inputTokens,
+    outputTokens,
+    totalTokens,
+});
+// 2.5 TTS: generateContent, direction embedded in the prompt, headerless PCM with the rate in the mimeType.
+const generateWithContent = async (ai, { model, text, voice, instructions }) => {
+    const response = await ai.models.generateContent({
+        model,
+        contents: [{ parts: [{ text: geminiDirectorsNotesPrompt(text, instructions) }] }],
+        config: {
+            responseModalities: ["AUDIO"],
+            speechConfig: { voiceConfig: { prebuiltVoiceConfig: { voiceName: voice } } },
+        },
+    });
+    const inlineData = response.candidates?.[0]?.content?.parts?.[0]?.inlineData;
+    const pcmBase64 = inlineData?.data;
+    if (!pcmBase64 || typeof pcmBase64 !== "string")
+        throw new Error("No audio data returned");
+    // Extract sample rate from mimeType (e.g., "audio/L16;codec=pcm;rate=24000")
+    const rateMatch = inlineData?.mimeType?.match(/rate=(\d+)/);
+    const sampleRate = rateMatch ? parseInt(rateMatch[1]) : GEMINI_TTS_SAMPLE_RATE;
+    const metadata = response.usageMetadata;
+    return {
+        rawPcm: Buffer.from(pcmBase64, "base64"),
+        sampleRate,
+        usage: metadata ? geminiUsage(model, metadata.promptTokenCount, metadata.candidatesTokenCount, metadata.totalTokenCount) : undefined,
+    };
+};
+const generateWithInteractions = async (ai, input) => {
+    // Passed per request too: the Interactions client is built apart from the models client.
+    const interaction = await ai.interactions.create(buildGeminiInteractionsTtsRequest(input), { timeout: GENAI_REQUEST_TIMEOUT_MS });
+    const usage = interaction.usage;
+    return {
+        ...pcmFromInteractionAudio(interaction.output_audio),
+        usage: usage ? geminiUsage(input.model, usage.total_input_tokens, usage.total_output_tokens, usage.total_tokens) : undefined,
+    };
 };
+export const generateGeminiTts = (ai, input) => usesGeminiInteractionsTts(input.model) ? generateWithInteractions(ai, input) : generateWithContent(ai, input);
 export const ttsGeminiAgent = async ({ namedInputs, params, config, }) => {
     const { text } = namedInputs;
     const { model, voice, suppressError, instructions } = params;
@@ -25,38 +60,8 @@ export const ttsGeminiAgent = async ({ namedInputs, params, config, }) => {
     const geminiResult = await (async () => {
         try {
             const ai = new GoogleGenAI({ apiKey, httpOptions: { timeout: GENAI_REQUEST_TIMEOUT_MS } });
-            const response = await ai.models.generateContent({
-                model: model ?? provider2TTSAgent.gemini.defaultModel,
-                contents: [{ parts: [{ text: getPrompt(text, instructions) }] }],
-                config: {
-                    responseModalities: ["AUDIO"],
-                    speechConfig: {
-                        voiceConfig: {
-                            prebuiltVoiceConfig: { voiceName: voice ?? provider2TTSAgent.gemini.defaultVoice },
-                        },
-                    },
-                },
-            });
-            const inlineData = response.candidates?.[0]?.content?.parts?.[0]?.inlineData;
-            const pcmBase64 = inlineData?.data;
-            const mimeType = inlineData?.mimeType;
-            if (!pcmBase64 || typeof pcmBase64 !== "string")
-                throw new Error("No audio data returned");
-            // Extract sample rate from mimeType (e.g., "audio/L16;codec=pcm;rate=24000")
-            const rateMatch = mimeType?.match(/rate=(\d+)/);
-            const sampleRate = rateMatch ? parseInt(rateMatch[1]) : 24000;
-            const rawPcm = Buffer.from(pcmBase64, "base64");
-            const usedModel = model ?? provider2TTSAgent.gemini.defaultModel;
-            const usage = response.usageMetadata
-                ? {
-                    provider: "gemini",
-                    model: usedModel,
-                    inputTokens: response.usageMetadata.promptTokenCount,
-                    outputTokens: response.usageMetadata.candidatesTokenCount,
-                    totalTokens: response.usageMetadata.totalTokenCount,
-                }
-                : undefined;
-            return { rawPcm, sampleRate, usage };
+            const request = { model: model ?? provider2TTSAgent.gemini.defaultModel, text, voice: voice ?? provider2TTSAgent.gemini.defaultVoice, instructions };
+            return await generateGeminiTts(ai, request);
         }
         catch (e) {
             if (suppressError) {

package/lib/data/scriptTemplates.js

@@ -484,7 +484,7 @@ export const scriptTemplates = [
         ],
         filename: "html",
         htmlImageParams: {
-            model: "claude-sonnet-4-5-20250929",
+            model: "claude-sonnet-4-6",
             provider: "anthropic",
         },
         lang: "en",
@@ -987,7 +987,7 @@ export const scriptTemplates = [
         ],
         filename: "presentation",
         htmlImageParams: {
-            model: "claude-sonnet-4-5-20250929",
+            model: "claude-sonnet-4-6",
             provider: "anthropic",
         },
         lang: "en",

package/lib/data/templateDataSet.js

@@ -49,7 +49,7 @@ export const templateDataSet = {
         "```",
     html: "Another LLM will generate actual slides from the prompt and data for each beat. Adding optional data would help it to generate more compelling slide. Mention the reference in one of beats, if it exists. The valid type of reference is 'article', 'paper', 'image', 'video', 'audio'. Use the JSON below as a template.\n" +
         "```JSON\n" +
-        '{"$mulmocast":{"version":"1.1","credit":"closing"},"references":[{"url":"https://www.somegreatwebsite.com/article/123","title":"Title of the article we are referencing","type":"[TYPE OF ARTICLE: article, paper, image, video, audio]"}],"title":"[TITLE: Brief, engaging title for the topic]","htmlImageParams":{"provider":"anthropic","model":"claude-sonnet-4-5-20250929"},"lang":"en","beats":[{"text":"[NARRATION: Narration for the beat.]","htmlPrompt":{"prompt":"[PROMPT to create appropriate HTML page for the beat.]"}},{"text":"[NARRATION: Narration for the beat.]","htmlPrompt":{"prompt":"[PROMPT to create appropriate HTML page for the beat with the data.]","data":{"description":"DATA TO BE PRESENTED IN THIS BEAT (in any format)]","net_income":{"Q2 FY2024":320,"Q3 FY2024":333,"Q4 FY2024":350},"unit":"USD (Million)"}}}],"canvasSize":{"width":1536,"height":1024}}\n' +
+        '{"$mulmocast":{"version":"1.1","credit":"closing"},"references":[{"url":"https://www.somegreatwebsite.com/article/123","title":"Title of the article we are referencing","type":"[TYPE OF ARTICLE: article, paper, image, video, audio]"}],"title":"[TITLE: Brief, engaging title for the topic]","htmlImageParams":{"provider":"anthropic","model":"claude-sonnet-4-6"},"lang":"en","beats":[{"text":"[NARRATION: Narration for the beat.]","htmlPrompt":{"prompt":"[PROMPT to create appropriate HTML page for the beat.]"}},{"text":"[NARRATION: Narration for the beat.]","htmlPrompt":{"prompt":"[PROMPT to create appropriate HTML page for the beat with the data.]","data":{"description":"DATA TO BE PRESENTED IN THIS BEAT (in any format)]","net_income":{"Q2 FY2024":320,"Q3 FY2024":333,"Q4 FY2024":350},"unit":"USD (Million)"}}}],"canvasSize":{"width":1536,"height":1024}}\n' +
         "```",
     image_prompt: "Another AI will generate images for each beat based on the image prompt of that beat. Movie prompts must be written in English. Mention the reference in one of beats, if it exists. Use the JSON below as a template.\n" +
         "```JSON\n" +

package/lib/methods/mulmo_presentation_style.js

@@ -8,7 +8,7 @@ import { userAssert } from "../utils/utils.js";
 import { slideThemes } from "../data/slideThemes.js";
 import { text2ImageProviderSchema, text2HtmlImageProviderSchema, text2MovieProviderSchema, text2SpeechProviderSchema, mulmoCanvasDimensionSchema, mulmoTransitionSchema, mulmoFillOptionSchema, } from "../types/schema.js";
 import { MulmoBeatMethods } from "./mulmo_beat.js";
-import { provider2ImageAgent, provider2MovieAgent, provider2LLMAgent, provider2TTSAgent, provider2SoundEffectAgent, provider2LipSyncAgent, defaultProviders, getModelAudio, AUDIO_MODE_ALWAYS, } from "../types/provider2agent.js";
+import { provider2ImageAgent, provider2MovieAgent, defaultMovieModel, provider2LLMAgent, provider2TTSAgent, provider2SoundEffectAgent, provider2LipSyncAgent, defaultProviders, getModelAudio, AUDIO_MODE_ALWAYS, } from "../types/provider2agent.js";
 const defaultTextSlideStyles = [
     '*,*::before,*::after{box-sizing:border-box}body,h1,h2,h3,h4,p,figure,blockquote,dl,dd{margin:0}ul[role="list"],ol[role="list"]{list-style:none}html:focus-within{scroll-behavior:smooth}body{min-height:100vh;text-rendering:optimizeSpeed;line-height:1.5}a:not([class]){text-decoration-skip-ink:auto}img,picture{max-width:100%;display:block}input,button,textarea,select{font:inherit}@media(prefers-reduced-motion:reduce){html:focus-within{scroll-behavior:auto}*,*::before,*::after{animation-duration:.01ms !important;animation-iteration-count:1 !important;transition-duration:.01ms !important;scroll-behavior:auto !important}}',
     "body { margin: 60px; margin-top: 40px; color:#333; background-color:#fff; font-size: 30px; font-family: Arial, sans-serif; box-sizing: border-box; height: 100vh }",
@@ -160,7 +160,7 @@ export const MulmoPresentationStyleMethods = {
         const imageParams = { ...presentationStyle.imageParams, ...beat?.imageParams };
         const provider = MulmoPresentationStyleMethods.getText2ImageProvider(imageParams?.provider) ?? defaultProviders.text2image;
         const agentInfo = provider2ImageAgent[provider];
-        // The default text2image model is gpt-image-1 from OpenAI.
+        // The default text2image model is gpt-image-2.5-sunburst from OpenAI.
         const defaultImageParams = {
             provider,
             model: agentInfo.defaultModel,
@@ -189,7 +189,7 @@ export const MulmoPresentationStyleMethods = {
             return true;
         }
         const provider = text2MovieProviderSchema.parse(movieParams?.provider ?? defaultProviders.text2movie);
-        const model = movieParams?.model ?? provider2MovieAgent[provider].defaultModel;
+        const model = movieParams?.model ?? defaultMovieModel(provider, movieParams);
         return getModelAudio(provider, model)?.mode === AUDIO_MODE_ALWAYS;
     },
     getSoundEffectAgentInfo(presentationStyle, beat) {

package/lib/types/provider2agent.d.ts

@@ -42,21 +42,25 @@ export declare const provider2TTSAgent: {
     };
 };
 export declare const gptImages: string[];
+export declare const isGptImage25Model: (model: string) => boolean;
+export declare const gptImageQuality: <Quality extends string>(model: string, quality?: Quality) => Quality | "high" | undefined;
 export declare const deprecatedOpenAIImageModelHints: {
-    readonly "dall-e-2": "Use 'gpt-image-1' or another supported model.";
-    readonly "dall-e-3": "Use 'gpt-image-1' or another supported model.";
+    readonly "dall-e-2": "Use 'gpt-image-2.5-sunburst' or another supported model.";
+    readonly "dall-e-3": "Use 'gpt-image-2.5-sunburst' or another supported model.";
 };
 export type DeprecatedOpenAIImageModel = keyof typeof deprecatedOpenAIImageModelHints;
 export declare const deprecatedGoogleImageModelHints: {
-    readonly "imagen-3.0-generate-002": "Use 'gemini-2.5-flash-image' or 'gemini-3-pro-image-preview' instead.";
-    readonly "imagen-4.0-generate-001": "Use 'gemini-2.5-flash-image' or 'gemini-3-pro-image-preview' instead.";
-    readonly "imagen-4.0-ultra-generate-001": "Use 'gemini-2.5-flash-image' or 'gemini-3-pro-image-preview' instead.";
-    readonly "imagen-4.0-fast-generate-001": "Use 'gemini-2.5-flash-image' or 'gemini-3-pro-image-preview' instead.";
-    readonly "imagen-4.0-generate-preview-06-06": "Use 'gemini-2.5-flash-image' or 'gemini-3-pro-image-preview' instead.";
-    readonly "imagen-4.0-ultra-generate-preview-06-06": "Use 'gemini-2.5-flash-image' or 'gemini-3-pro-image-preview' instead.";
+    readonly "imagen-3.0-generate-002": "Use 'gemini-3.1-flash-lite-image' or 'gemini-3-pro-image' instead.";
+    readonly "imagen-4.0-generate-001": "Use 'gemini-3.1-flash-lite-image' or 'gemini-3-pro-image' instead.";
+    readonly "imagen-4.0-ultra-generate-001": "Use 'gemini-3.1-flash-lite-image' or 'gemini-3-pro-image' instead.";
+    readonly "imagen-4.0-fast-generate-001": "Use 'gemini-3.1-flash-lite-image' or 'gemini-3-pro-image' instead.";
+    readonly "imagen-4.0-generate-preview-06-06": "Use 'gemini-3.1-flash-lite-image' or 'gemini-3-pro-image' instead.";
+    readonly "imagen-4.0-ultra-generate-preview-06-06": "Use 'gemini-3.1-flash-lite-image' or 'gemini-3-pro-image' instead.";
 };
 export type DeprecatedGoogleImageModel = keyof typeof deprecatedGoogleImageModelHints;
+export declare const isGeminiImageModel: (model: string) => boolean;
 export declare const vertexAIGlobalOnlyImageModels: ReadonlySet<string>;
+export declare const vertexImageLocation: (model: string, location?: string) => string;
 export declare const provider2ImageAgent: {
     openai: {
         agentName: string;
@@ -169,34 +173,34 @@ export declare const provider2LipSyncAgent: {
 export declare const provider2LLMAgent: {
     readonly openai: {
         readonly agentName: "openAIAgent";
-        readonly defaultModel: "gpt-5";
+        readonly defaultModel: "gpt-5.6-sol";
         readonly keyName: "OPENAI_API_KEY";
         readonly baseURLKeyName: "OPENAI_BASE_URL";
         readonly apiVersionKeyName: "OPENAI_API_VERSION";
         readonly max_tokens: 8192;
-        readonly models: readonly ["gpt-5", "gpt-5-nano", "gpt-5-mini", "gpt-4.1", "gpt-4.1-mini", "gpt-4.1-nano", "o3", "o3-mini", "o3-pro", "o1", "o1-pro", "gpt-4o", "gpt-4o-mini"];
+        readonly models: readonly ["gpt-5.6-sol", "gpt-5.6-terra", "gpt-5.6-luna", "gpt-5", "gpt-5-nano", "gpt-5-mini", "gpt-4.1", "gpt-4.1-mini", "gpt-4.1-nano", "o3", "o3-mini", "o3-pro", "o1", "o1-pro", "gpt-4o", "gpt-4o-mini"];
     };
     readonly anthropic: {
         readonly agentName: "anthropicAgent";
-        readonly defaultModel: "claude-sonnet-4-5-20250929";
+        readonly defaultModel: "claude-sonnet-4-6";
         readonly max_tokens: 8192;
-        readonly models: readonly ["claude-opus-4-1-20250805", "claude-opus-4-20250514", "claude-sonnet-4-20250514", "claude-sonnet-4-5-20250929", "claude-haiku-4-5-20251001"];
+        readonly models: readonly ["claude-sonnet-4-6", "claude-sonnet-4-5-20250929", "claude-haiku-4-5-20251001"];
         readonly keyName: "ANTHROPIC_API_KEY";
         readonly apiKeyNameOverride: "ANTHROPIC_API_TOKEN";
     };
     readonly gemini: {
         readonly agentName: "geminiAgent";
-        readonly defaultModel: "gemini-2.5-flash";
+        readonly defaultModel: "gemini-3.8-flash";
         readonly max_tokens: 8192;
-        readonly models: readonly ["gemini-3.1-pro-preview", "gemini-3-flash-preview", "gemini-2.5-pro", "gemini-2.5-flash", "gemini-2.5-flash-lite"];
+        readonly models: readonly ["gemini-3.8-flash", "gemini-3.1-pro-preview", "gemini-3-flash-preview", "gemini-2.5-pro", "gemini-2.5-flash", "gemini-2.5-flash-lite"];
         readonly keyName: "GEMINI_API_KEY";
     };
     readonly groq: {
         readonly agentName: "groqAgent";
-        readonly defaultModel: "llama-3.1-8b-instant";
+        readonly defaultModel: "openai/gpt-oss-20b";
         readonly keyName: "GROQ_API_KEY";
         readonly max_tokens: 4096;
-        readonly models: readonly ["llama-3.1-8b-instant", "llama-3.3-70b-versatile", "deepseek-r1-distill-llama-70b", "openai/gpt-oss-120b", "openai/gpt-oss-20b"];
+        readonly models: readonly ["openai/gpt-oss-20b", "openai/gpt-oss-120b"];
     };
     readonly mock: {
         readonly agentName: "mediaMockAgent";
@@ -229,6 +233,13 @@ export declare const ProviderName: {
 export declare const isImageProvider: (provider: string | undefined) => provider is keyof typeof provider2ImageAgent;
 export declare const getMaxImageReferenceImages: (provider: string | undefined, model: string) => number | undefined;
 export declare const getModelAudio: (provider: keyof typeof provider2MovieAgent, model: string) => MovieAudioSpec | undefined;
+export declare const defaultMovieModel: (provider: keyof typeof provider2MovieAgent, movieParams?: {
+    vertexai_project?: string;
+}) => string;
+export declare const VEO_EXTENSION_MODEL = "veo-3.1-generate-preview";
+export declare const VEO_SEGMENT_SEC = 8;
+export declare const veoExtensionCount: (requestedSec: number) => number;
+export declare const veoExtendedSeconds: (requestedSec: number) => number;
 export declare const getModelDuration: (provider: keyof typeof provider2MovieAgent, model: string, movieDuration?: number) => number | undefined;
 export type ModelPricing = {
     unit: "tokens" | "chars" | "seconds" | "images";